robust and reversible audio watermarking by modifying...
TRANSCRIPT
Research ArticleRobust and Reversible Audio Watermarking by ModifyingStatistical Features in Time Domain
Shijun Xiang12 Le Yang1 and Yi Wang1
1The School of Information Science and Technology Jinan University Guangzhou 510632 China2State Key Laboratory of Information Security Institute of Information Engineering The Chinese Academy of SciencesBeijing 100093 China
Correspondence should be addressed to Shijun Xiang shijun_xiangqqcom
Received 10 January 2017 Revised 31 March 2017 Accepted 4 April 2017 Published 27 April 2017
Academic Editor AkramM Z M Khedher
Copyright copy 2017 Shijun Xiang et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
Robust and reversible watermarking is a potential technique in many sensitive applications such as lossless audio or medical imagesystems This paper presents a novel robust reversible audio watermarking method by modifying the statistic features in timedomain in the way that the histogram of these statistical values is shifted for data hiding Firstly the original audio is divided intononoverlapped equal-sized frames In each frame the use of three samples as a group generates a prediction error and a statisticalfeature value is calculated as the sum of all the prediction errors in the frameThe watermark bits are embedded into the frames byshifting the histogram of the statistical features The watermark is reversible and robust to common signal processing operationsExperimental results have shown that the proposed method not only is reversible but also achieves satisfactory robustness to MP3compression of 64 kbps and additive Gaussian noise of 35 dB
1 Introduction
With the rapid development of the Internet technologypublication and dissemination of digital multimedia becomemore and more convenient However the authenticity andsecurity of the digital multimedia are a challenge for themedia owner [1] Digital watermarking technology is anefficient approach to protect the copyright of the digitalmedia Reversible watermarking is one of the watermarkingtechnologies used for data hiding Reversible watermarkingenables embedding secret data into host media and allowextraction of the original media and the secret data [2ndash4] Itis very useful in sensitive applications such as medical imagesystemmilitary image and lossless audio [5] Although thereare so many reversible watermarking methods most of themare designed in a lossless environment and cannot resist anytype of attacks As a result the original media or the secretdata cannot be recovered after the watermarked media gothrough some changes [6]
In some cases such as the copyright protection of thedigital media the embedded data is expected to be robust tosome attacks such as lossy compression or additive noise Tothis end researchers pay more attention to robust reversible
watermarking Robust reversible watermarking is that theoriginal media and the embedded data can be both recoveredcorrectlywhen thewatermarkedmedia remain intact and theembedded data can still be extractedwithout error evenwhenthe watermarked media go through some attacks [7] Untilnow a few robust reversible image watermarking methodshave been proposed which can be classified into two groups
(i) Blind watermarking scheme in [7 8] Vleeschouweret al proposed a blind extraction scheme based on thepatchwork theory andmodulo-256 by using the gray-scale histogram rotation This work is robust againstJPEG compression but the watermarked image haslower visible quality due to the reason that thewatermark embedding procedure will cause salt-and-pepper noise in the watermarked image Besides thepayload is low To handle the salt-and-pepper noiseproblem Zou et al proposed a scheme by shifting theabsolute mean values of the integer wavelet transform(IWT) coefficients in a chosen subband [9] and Niet al proposed a scheme by modifying the histogramof a robust statistical quantity in the spatial domain[10] Since the embedding process may introduce the
HindawiAdvances in MultimediaVolume 2017 Article ID 8492672 10 pageshttpsdoiorg10115520178492672
2 Advances in Multimedia
error bits the error correction coding (ECC) hasbeen used Besides these two methods suffered fromthe unstable robustness and incomplete reversibilityaccording to [11] In [12] Zeng et al enhanced thescheme of Ni et al by introducing two thresholds anda new embedding mechanism This method is blindand reversible For a satisfactory performance thetwo threshold values have to be carefully searched fordifferent cover images
(ii) Nonblind watermarking scheme in [13] a nonblindscheme based on wavelet-domain statistical quantityhistogram shifting and clustering (WSQH-SC) isproposed A pixel adjustment is presented at first toavoid the overflow and underflow and a locationmapis used to record the changed pixels This methodachieved good robustness against JPEG JPEG2000and additive Gaussian noise but it is not blind sincethe locations of the changed pixels need to be savedas a part of side information and transmitted to thereceiver side in order to recover the original imageIn [14] the Slantlet transform (SLT) was applied toimage blocks modifying the mean values of the HLand LH subband coefficients to embed the watermarkbits and a second stage of SLT transform is appliedto the LL1 subband embedding another watermarkbit into the HL2 and LH2 subband Because thecoefficients and the mean values are fractional withmore decimal places themean informationwas takenas side information to be sent to the receiver side forthe recovery of the original cover image In order tosolve the nonblind extraction question in [14] theauthors in [15] used IWT on images and randomlyselected 10 coefficients from all the 16 coefficients ina block to compute the amplitude mean of the blockso that the mean information can be embedded intothe image itself for blind extraction
In [16] Coltuc and Chassery proposed a general frame-work for robust reversible watermarking by multiple water-marking First the watermark is embedded into the coverimage with a robust watermarking method and then areversible watermarking method is adopted to embed theinformation (which is used to restore the original coverimage) into the robust watermarked image Suppose 119868 and1198681 are the original image and the robust watermarked imageafter embedding a watermark119908 respectivelyThe embeddingdistortion 119889 = 119868 minus 1198681 is compressed and embedded into therobust watermarked image with the reversible watermarkingmethod At the receiver side if there are no attacks the robustwatermarked image 1198681 and the difference 119889 can be extractedsince the embedding process is reversible Then the originalimage 119868 can be recovered by 119868 = 119889 + 1198681 Furthermorethe watermark can be extracted If the watermarked imagegoes through a JPEG compression operation the robustwatermark can still be extracted This framework is veryinstructive and achieves higher payload and good robustnessagainst JPEG compression
In [17] a robust reversible audio method based on spreadspectrum and amplitude expansion is proposed A robust
payload is embedded at first using the direct-sequencespread-spectrummodulation with the sequence determinedfrom the amplitude expansion in time and frequency of inte-ger modified discrete cosine transform (MDCT) coefficientsThen a reversible payload is embedded into the apertures inthe amplitude histogram that result from amplitude expan-sion of the integer MDCT coefficients to recover the hostaudio This method achieves robustness against some signalprocessing like MP3 compression and additive noise and ifthe watermarked audio remains intact the host audio can berecovered perfectly
In this paper we propose a novel robust and reversibleaudio watermarking scheme based on statistic feature andhistogram shifting in time domain By shifting the histogramof the statistic features in time domain the proposed algo-rithm achieves good robustness and reversibility at the sametime
The rest of the paper is organized as follows The foun-dation work is introduced in Section 2 The proposed water-marking algorithm is described in Section 3 Experimentalresults are presented in Section 4 Section 5 concludes thispaper
2 Algorithmrsquos Principle
This section will introduce the foundation works of the pro-posed robust reversible digital audio watermarking schemeFirstly a robust statistic feature of time domain is introducedthen how to modify the statistic feature to embed thewatermark bit is briefly described
21 Robust Statistic Feature Consider a time-discrete digitalaudio signal 119883 the host signal is first divided into nonover-lapped equal-sized frames We take119873 samples per frame forexample119873 samples as a frame and three samples as a groupare shown in Figure 1 For a sample group (119909119897 119909119898 and 119909119903)the prediction value of the middle sample 119909119898 is calculated byusing two immediate samples as
119909119898 = lceil119909119897 + 1199091199032 rceil (1)
where lceil119909rceil means rounding the elements of 119909 to the nearestinteger towards infinity The prediction error of 119909119898 is
119890119898 = 119909119898 minus 119909119898 (2)
Since the samples in a group are often highly correlatedthe prediction error 119890119898 is expected to be very close to zeroFor a frame with 119873 samples 1198733 prediction errors can becomputed The sum of all the prediction errors in a framedenoted by 119864 is called the statistic feature in this paper Thestatistical feature of a frame is calculated as
119864 =1198733
sum119894=1
119890119894 (3)
where 119890119894 is the prediction error of the 119894th group in the frameThe basic idea of the proposed algorithm is based on thisstatistic property
Advances in Multimedia 3
x1 x2 x3 x4 x5 x6 middot middot middot XNminus2 XNminus1 xN
Group 1 Group 2 middot middot middot Group i
Figure 1 The use of 119873 samples as a frame and three samples as agroup
3002001000 400 500minus200minus300 minus100minus4000
2
4
6
8
10
12
14
Figure 2 The distribution of the 119864 values for the clip track 1
22 Watermark Statistic Feature For each frame one water-mark bit is embedded by shifting the value of the statisticfeature The shifting operation is done by modifying thesamples in a frame Taking track 1 (which is downloadedfrom the website [18]) as example clip Figure 2 shows thedistribution of 119864 values by using 300 samples as a frame andthree samples as a group The rule to modify the statisticvalue is referred to histogram shifting method At first wescan all frames and find out the maximum of the absolute119864 values denoted by 119864max Then a threshold 119879 is set to apositive integer bigger than 119864max As a result all 119864 values arewithin the range [minus119879 119879] For example from Figure 2 we canget that 119864max is 446 so threshold 119879 can be an integer suchas 500 The watermarking rule is to keep the statistic featurewithin [minus119879 119879] if the watermark bit is ldquo0rdquo while the statisticfeature is shifted away from zero by a shifting quantity 119879 + 119866if the watermark bit is ldquo1rdquo To achieve stronger robustnessparameter 119866 is a threshold which is usually set bigger than119879 To reduce the embedding distortion if the embeddedwatermark bit is ldquo1rdquo and the original statistic feature belongsto [0 119879) the statistic feature is shifted to the region [119879 +119866 2119879 + 119866] if the embedded watermark bit is ldquo1rdquo and theoriginal statistic feature belongs to (minus119879 0) the statistic featureis shifted to the region [minus2119879 minus 119866 minus119879 minus 119866] In such a waythe bit-0 region and the bit-1 region are separated by therobust regions (119879 119879 + 119866) and (minus119879 minus 119866 minus119879) For exampleFigure 3 shows the distribution of 119864 values after embeddingwatermark by using clip track 1
The modifying rules are as followsIf the embedded bit is ldquo0rdquo keep the frame unchanged If
the embedded bit is ldquo1rdquo the samples in the frame aremodifiedby
2T
Bit-1region
Robustregion
Bit-0region
Robustregion
Bit-1region
GG
0
5
10
15
20
25
30
35
200010000 3000 4000minus2000minus3000 minus1000minus4000
Figure 3The distribution of the 119864 values of track 1 after embeddingwatermark
119910119896119894 =
119909119896119894 + 120573 if 0 le 119864119896 lt 119879 mod (119894 3) = 2119909119896119894 minus 120573 if minus 119879 lt 119864119896 lt 0 mod (119894 3) = 2119909119896119894 otherwise
(4)
where 119909119896119894 is the 119894th sample in the 119896th frame The index 119894 isin [1 119878] and 119878 is the number of the samples in a frame Theinteger value 120573 is the shifted quantity of a sample
120573 = lceil119879 + 1198661198783 rceil (5)
At the receiver side if the watermarked audio remainsintact the watermark bits can be extracted by
119908119896 = 0 if 119864119896 isin [minus119879 119879]1 otherwise (6)
where 119908119896 is the to be hidden 119896th bit The original audio canbe recovered as
119909119896119894
=
119910119896119894 minus 120573 if119864119896 isin [119879 + 119866 2119879 + 119866] mod (119894 3) = 2119910119896119894 + 120573 if119864119896 isin [minus2119879 minus 119866 minus119879 minus 119866] mod (119894 3) = 2119910119896119894 otherwise
(7)
23 Prevention of OverflowUnderflow For a 16-bit digitalaudio the permission range of the sample value is [minus215 215]Watermark embedding will modify the sample values withthe value 120573 so the overflow or underflow does not occur ifthe original sample values belong to [minus215 + 120573 215 minus 120573] Infact as the value 120573 is very small the original sample values ofmost normal audio belong to [minus215 + 120573 215 minus 120573] Thereforein the proposedmethod there is no overflow or underflow inmost cases If the audio cannot meet this condition we canrecord the location and modify the sample value to the range[minus215 + 120573 215 minus 120573] then the location can be saved as sideinformation and embedded into the audio
4 Advances in Multimedia
Input audio X andwatermark w
Divide audio into framessized S samples
Calculate the statisticfeatures value E of the
frames
Set threshold T(T gt max(abs(E))) and
threshold G (usually G gt T)
wk = 1
No change
Combine the frames to obtainthe watermarked audio
End
Ek isin [0 T]
Yes
Yes
Some samples in the frameare added by value 훽 and
E㰀k = Ek + T + G
No
No
Some samples in the frameare subtracted by value 훽 and
E㰀k = Ek minus T minus G
Figure 4 Watermark embedding process
3 Proposed Algorithm
The embedding and extraction processes are presented indetail as follows
31 Watermark Embedding Figure 4 shows the proposedwatermark embedding process The watermark is embeddedwith the following five steps
Step 1 Divide the original audio 119883 into nonoverlappingframes sized 119878 samples
Step 2 Calculate the statistic features of the frames (119864 values)by referring to (1)ndash(3)
Step 3 Set the threshold values 119879 and 119866 (119879 gt 119864max andusually 119866 gt 119879)Step 4 If the watermark bit is ldquo0rdquo nothing is changed Ifthe bit is ldquo1rdquo shift the statistic feature value with a shiftingquantity 119879 +119866 to embed the watermark bit by modifying thesamples in the frame with value 120573 referring to (4)Step 5 Combine the frames to get the watermarked audio
32 Watermark Extraction If the watermarked audio goesthrough some attacks (such as MP3 compression additivenoise resampling or requantization) the watermark can
still be detected To improve the accuracy of the watermarkextraction three extraction methods and a majority votingsystem are adopted to identify the extracted watermark bycomputing the distorted statistical feature values 1198641015840(i) Extraction 1 Redefine the bit-0 region as [minus119879 minus 1198662 119879 +1198662] and the watermark extraction as
11990810158401 (119894) = 0 if1198641015840 isin [minus119879 minus 1198662 119879 + 1198662]1 otherwise (8)
(ii) Extraction 2 Redefine the bit-0 region as [minus119879 minus 1198663 119879 +1198663] and the watermark extraction as
11990810158402 (119894) = 0 if1198641015840 isin [minus119879 minus 1198663 119879 + 1198663]1 otherwise (9)
(iii) Extraction 3 119896-means clustering algorithm is introducedto extract bits Figure 5 shows the distribution of the 119864 valuesafter MP3 compression and the watermark can be extractedby
11990810158403 (119894) = 0 if1198641015840 isin class 21 if1198641015840 isin class 1 and class 3 (10)
Advances in Multimedia 5
Class 1 Class 3
Class 2
3000
2000
10000
4000
5000
minus20
00
minus30
00
minus40
00
minus10
00
minus50
00
02468
101214161820
Figure 5 The distribution of the statistic feature values of track 1after MP3 compression at 64 kbps
The majority voting system works as
1199081015840 (119894) =
11990810158401 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158401 (119894) = 11990810158403 (119894)11990810158401 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158401 (119894) = 11990810158403 (119894)11990810158402 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158402 (119894) = 11990810158403 (119894)11990810158403 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158401 (119894) = 11990810158403 (119894)
(11)
Eventually three extraction methods and a majorityvoting system are adopted to extract watermark Figure 6shows the proposed watermark extracting process If thewatermarked audio remains intact the watermark can beextracted correctly and the original audio can be recoveredas the following steps
Step 1 Divide the watermarked audio 119883119908 into nonoverlap-ping frames sized 119878 samples
Step 2 Calculate the statistic feature of the frames (1198641015840 values)by referring to (1)ndash(3)
Step 3 Extract the watermark with three extraction methodsand identify the watermark with the majority voting systemby referring to (8)ndash(11)
Step 4 The original audio can be recovered by modifying thesamples in the frame with value 120573 referring to (7)Step 5 Combine the frames to get the original audio
If the watermarked audio goes through some attacks theoriginal audio cannot be recovered exactly so we focus onthe watermark extraction and the watermark is extracted asfollows
Step 1 Divide the watermarked audio 119883119908 into nonoverlap-ping frames sized 119878 samples
Step 2 Calculate the statistic feature values of the frames 1198641015840by referring to (1)ndash(3)
Step 3 Extract the watermark with three extraction methodsand identify the watermark with the majority voting systemby referring to (8)ndash(11)
4 Experimental Results
In this section 7WAVaudio file of the sample rate of 441 KHzand 16 bits per sample (tracks 1 2 3 4 5 6 and 7 [18]) are usedas example clips to evaluate the performance of the proposedalgorithm The payload of our method only depends on thelength of a frame 119878 for a time-discrete digital audio signal119883in length119872 the pure payload can be calculated by
119862 = lfloor119872119878 rfloor (12)
In the experiment the watermark is a pseudo-randomsequence in length of 1000 bits The imperceptibility is firstanalyzed by the SNR standard at different threshold valuesand different sample numbers per frame Then robustnesstesting against MP3 compression additive noise (AWGN)resampling (441-16-441 kHz) and requantization (16-8-16bits) are reported by using the software CoolEditPro v21
41 Imperceptibility Test The imperceptibility is measuredby the embedding distortion In the proposed scheme thedistortion is caused by the shifting quantity on the samplesdepending on thresholds 119879 119866 and the length of a frame 119878Since 119879 is set at first we only investigate the influence of 119866and 119878 on SNR
Figure 7 plots the relationship between SNR and thethreshold 119866 for different clips at the same threshold 119879 and119878 From this figure we can conclude that with the increase of119866 the SNR value drops The reason is that the larger 119866 isthe larger shifting quantity is used so the larger embeddingdistortion is caused As a result SNR value drops
Figure 8 plots the relationship between SNR and thelength of a frame 119878 for different clips at the same thresholds119879 and 119866 We can see from this figure that the larger 119878is the higher SNR value is achieved The reason is thatwith the increase of 119878 the shifting quantity for every singlesample drops so the SNR values rise due to the fact that theembedding distortion is reduced Consequently the framelength 119878 will influence Maximum Embedding Capacity andSNR value directly the Maximum Embedding Capacity ishigher when 119878 is smaller according to (12) and the SNR valueis higher when 119878 is larger according to Figure 8 To considerthe balance between Maximum Embedding Capacity andSNR value we have found that the value of 119878within the rangeof 300 to 600 can usually achieve a satisfactory result after aset of experiments
42 Robustness Testing To test the robustness of the pro-posed scheme a set of experiments has been taken ontracks 1ndash7 Table 1 shows the results in which RP meansresampling (441-16-441 kHz) operations while RQ meansrequantization (16-8-16 bits) operations We can observefrom this table that all the example clips can achieve therobustness against MP3 compression at 64Kbps For track1 the watermark bits can be correctly extracted under the
6 Advances in Multimedia
Input audio Xw
Divide audio into framessized S samples
Calculate the statisticfeatures value E㰀 of the
frames
Three extraction methodsand a majority voting system
Watermark w
End
Yes
Yes
No
No
E㰀k isin [T + G 2T + G]
No change
E㰀k isin [minus2T minus G minusT minus G]
Combine the frames toobtain the original audio
Some samples in the frameare subtracted by value 훽 and
Ek = E㰀k minus T minus G
Some samples in the frameare added by value 훽 and
Ek = E㰀k + T + G
Figure 6 Watermark extracting process
3500 4000 4500 5000 5500 6000 6500 7000 75003000Threshold G
40
45
50
55
60
SNR
(dB)
Track 1 T = 500 S = 300Track 2 T = 5300 S = 300Track 3 T = 300 S = 300Track 4 T = 9900 S = 300
Track 5 T = 200 S = 300Track 6 T = 200 S = 300Track 7 T = 1700 S = 300
Figure 7 Relationship between SNR and threshold 119866
MP3 compression of 48KbpsThe robustness against additivenoise is also satisfactory Even with the noise intensity at25 dB the BER (bit error rate) values are less than 10except for track 1 Besides the watermark robustness againstresampling and requantization operations is perfect and thehidden bits can be recovered without errors
350 400 450 500 550 600300Length S
1015202530354045505560
SNR
(dB)
Track 1 T = 550G = 3000Track 2 T = 6200G = 7000Track 3 T = 1700G = 4000Track 4 T = 16000G = 17000
Track 5 T = 300G = 3000Track 6 T = 200G = 3000Track 7 T = 2700G = 3000
Figure 8 Relationship between SNR and the frame length 119878
As shown in Figure 3 the robustness of the proposedmethod is originated from the robust region The robustregion depends on threshold 119866 The larger 119866 is the largerthe robust region is and the stronger robustness is Figure 9
Advances in Multimedia 7
Table 1 Performance of the proposed method
Audio 119878 119879 119866 SNR MP3 (Kbps) AWGN (35 dB) AWGN (30 dB) AWGN (25 dB) RP RQTrack 1 420 500 3000 5474 48 21000 771000 2511000 01000 01000Track 2 600 6200 7000 4798 64 01000 01000 01000 01000 01000Track 3 510 600 3000 5409 48 01000 01000 441000 01000 01000Track 4 300 9900 10000 3899 64 01000 01000 01000 01000 01000Track 5 300 200 3000 5618 48 01000 251000 771000 01000 01000Track 6 600 200 3000 495 48 01000 01000 461000 01000 01000Track 7 300 1600 4000 4378 48 01000 01000 01000 01000 01000
Table 2 Performance of the proposed method with different length of a frame 119878 on track 1
Track S T G MP3 AWGN AWGN AWGN(Kbps) (35 dB) (30 dB) (25 dB)
Track 1 300 500 3000 80 91000 901000 2991000Track 1 420 500 3000 56 121000 1711000 3201000Track 1 510 550 3000 80 281000 1711000 1581000Track 1 600 550 3000 48 281000 2211000 3901000Track 6 300 200 3000 48 01000 01000 31000Track 6 420 200 3000 56 281000 01000 231000Track 7 300 200 4000 48 01000 01000 01000Track 7 420 200 4000 56 01000 01000 01000
Track 1 G = 5000Track 1 G = 6000Track 2 G = 6000Track 2 G = 7000
Track 3 G = 3000Track 3 G = 4000Track 4 G = 9000Track 4 G = 10000
50 55 60 65 70 75 80 85 90 95 10045MP3 compression bit rate
0
001
002
003
004
005
Bit e
rror
rate
(BER
)
Figure 9 Relationship between MP3 bit rate and threshold 119866
supports the conclusion Figure 9 shows the bit error rate(BER) at different threshold 119866 for the same audio withsame threshold 119879 The lower BER means that the strongerrobustness is achieved We can find that as threshold 119866increases the BER drops and the robustness rises
Take track 1 as example clip Figure 10 shows the bit errorrate of the extracted watermark with different thresholds 119866against additive noises with the same 119879 and 119878 (119879 = 500 119878 =300) We can see that the larger 119866 is the smaller bit errorrate is and the better robustness is As threshold 119866 increasesthe robustness becomes stronger In the application we canadjust the parameter 119866 to achieve ideal robustness On the
25 30 35 4020AWGN (dB)
0005
01015
02025
03035
04045
05
Bit e
rror
rate
(BER
)
Track 1 G = 3000Track 1 G = 4000Track 1 G = 5000
Track 1 G = 6000Track 1 G = 7500
Figure 10 The BER values at different AWGN with differentthreshold 119866
other hand with the increase of 119866 the SNR value drops Toconsider the balance between SNR value and robustness wehave found that the value of119866within 3000 to 5000 can usuallyachieve a satisfactory result after a set of experiments
To evaluate the effect of the frame length 119878 on therobustness performance a set of experiments has been takenon track 1 track 6 and track 7 Table 2 lists the results Wecan observe that for the same audio with the same 119879 and 119866such as 119879 = 500 and 119866 = 3000 for track 1 as 119878 increasesthe robustness against MP3 compression will strengthen butfor track 6 and track 7 as 119878 increases the robustness against
8 Advances in Multimedia
Table3Com
paris
onwith
metho
din
[17]
Track
Metho
din
[17]
Prop
osed
metho
dPayload
ODG
MP3
AWGN
119878119879
119866Payload
ODG
MP3
MP3
AWGN
(bits)
(128
Kbps)
(36d
B)(bits)
(80K
bps)
(64K
bps)
(36d
B)Track32
216
minus245
00
150
1500
3000
1000
minus137
002
0Track35
216
minus131
012
150
3500
5000
1000
minus158
008
0Track65
216
minus121
06
150
302000
1000
minus023
001
12
Track66
216
minus024
06
9660
0060
001000
minus105
03
12
0Track69
216
minus031
14
300
400
2000
1000
minus023
00
0
Advances in Multimedia 9
MP3 compression drops so the effect of the frame length 119878on the robustness against MP3 compression is unstable Theinfluence on AWGN is little
For fair comparison with the method in [17] we use thesame host signals (tracks 32 35 65 66 and 69) downloadedfrom sound quality assessment material (SQAM) collection[19] Table 3 shows the robustness testing results against MP3compression and additive noise (AWGN) operations We canobserve that the method in [17] can carry 216 bits and resisttheMP3 compression at 128Kbpswhile the proposedmethodcan resist the MP3 compression at 64Kbps with 1000 bitsembedded In addition in our method the BER under theAWGN of 35 dB is less than that of the method in [17]In other words the proposed method can provide largerembedding capacity and obtain stronger robustness againstMP3 compression and AWGN attacks The imperceptibilityis evaluated by using the ODG standardThe closer the ODGvalue to 0 the better the imperceptibility For the table it isnoted that the imperceptibility of the proposed method isbetter except for the clips track 35 and track 66 The reason isthat119864max values of track 35 and track 66 are bigger As a resultthresholds 119879 and 119866 are also larger and more embeddingdistortion will be caused
5 Conclusions
In this paper we proposed a robust and reversible audiowatermarking method by shifting the histogram of thestatistical feature values in time domainThe statistical featureis calculated as the sum of the prediction errors in a frameSince the audio clip has a larger number of samples andeach frame can hold enough elements the statistical featureis robust to those common signal processing operationsConsidering that the distribution of the statistical featurevalues may be distorted to some extent three extractionmethods and the majority voting system are designed forthe watermark detection Experimental results have shownthat thousands of bits can be reversibly embedded and thewatermark bits can resist MP3 compression of 64 kbps andadditive noise of 35 dB Comparingwith the existing excellentmethod the proposed method can embed more watermarkbits and achieve stronger robustness
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This work was supported by NSFC (no 61272414) and OpenResearch Fund from the State Key Laboratory of InformationSecurity (no 2016-MS-07)
References
[1] Y Xiang I Natgunanathan S GuoW Zhou and S NahavandildquoPatchwork-based audio watermarking method robust to de-synchronization attacksrdquo IEEE Transactions on Audio Speechand Language Processing vol 22 no 9 pp 1413ndash1423 2014
[2] J Fridrich M Goljan and R Du ldquoLossless data embeddingmdashnew paradigm in digital watermarkingrdquo EURASIP Journal onAdvances in Signal Processing vol 2002 no 2 pp 185ndash196 2002
[3] Y Q Shi Z Ni D Zou C Liang and G Xuan ldquoLosslessdata hiding fundamentals algorithms and applicationsrdquo inProceeding of the IEEE International Symposium on Circuits andSystems vol 2 pp 313ndash336 2004
[4] S Lee CD Yoo andTKalker ldquoReversible imagewatermarkingbased on integer-to-integer wavelet transformrdquo IEEE Transac-tions on Information Forensics and Security vol 2 no 3 pp 321ndash330 2010
[5] X Li B Yang and T Zeng ldquoEfficient reversible watermarkingbased on adaptive prediction-error expansion and pixel selec-tionrdquo IEEE Transactions on Image Processing vol 20 no 12 pp3524ndash3533 2000
[6] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hidingrdquo in Proceeding of the IEEE Interna-tional Conference onMultimedia and Expo (ICME rsquo2004) vol 3pp 2199ndash2202 Taipei Taiwan June 2004
[7] C De Vleeschouwer J Delaigle and B Macq ldquoCircularinterpretation of histogram for reversible watermarkingrdquo inProceeding of the IEEE 4th Workshop on Multimedia SignalProcessing pp 345ndash350 Cannes France 2001
[8] C De Vleeschouwer J F Delaigle and B Macq ldquoCircularinterpretation of bijective transformations in lossless water-marking for media asset managementrdquo IEEE Transactions onMultimedia vol 5 no 1 pp 97ndash105 2003
[9] D Zou YQ Shi ZNi andW Su ldquoA semi-fragile lossless digitalwatermarking scheme based on integer wavelet transformrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 10 pp 1294ndash1300 2006
[10] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hiding designed for semi-fragile imageauthenticationrdquo IEEE Transactions on Circuits and Systems forVideo Technology vol 18 no 4 pp 890ndash896 2008
[11] L An X Gao C Deng and F Ji ldquoRobust lossless data hidingAnalysis and evaluationrdquo in Procceding of the InternationalConference on High Performance Computing and Simulation(HPCS rsquo10) pp 512ndash516 July 2010
[12] X-T Zeng L-D Ping and X-Z Pan ldquoA lossless robust datahiding schemerdquo Pattern Recognition vol 43 no 4 pp 1656ndash1667 2010
[13] L An X Gao X Li D Tao C Deng and J Li ldquoRobustreversible watermarking via clustering and enhanced pixel-wisemaskingrdquo IEEE Transactions on Image Processing vol 21 no 8pp 3598ndash3611 2012
[14] R Thabit and B E Khoo ldquoCapacity improved robust losslessimage watermarkingrdquo IET Image Processing vol 8 no 11 pp662ndash670 2014
[15] S Xiang and Y Wang ldquoDistortion-free robust reversible water-marking by modifying and recording iwt means of imageblocksrdquo in Proceeding of the 14th International Workshop(IWDW rsquo15) Tokyo Japan October 2015
[16] D Coltuc and J Chassery ldquoDistortion-free robust watermark-ing a case studyrdquo in Security Steganography andWatermarkingof Multimedia Contents vol 6505 of Proceedings of SPIE pp588ndash595 San Jose Calif USA 2007
[17] A Nishimura ldquoReversible and robust audio watermarkingbased on spread spectrum and amplitude expansionrdquo in Inter-national Workshop on Digital Watermarking (IWDW rsquo14) vol9023 of Lecture Notes in Computer Science pp 215ndash229
10 Advances in Multimedia
[18] Massachusetts Institute of Technology (MIT) Audio Databasehttpsoundmediamitedumediaphp
[19] EBUCommittee Sound quality assessmentmaterial recordingsfor subjective tests httpstechebuchpublicationssqamcd
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal of
Volume 201
Submit your manuscripts athttpswwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 201
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
2 Advances in Multimedia
error bits the error correction coding (ECC) hasbeen used Besides these two methods suffered fromthe unstable robustness and incomplete reversibilityaccording to [11] In [12] Zeng et al enhanced thescheme of Ni et al by introducing two thresholds anda new embedding mechanism This method is blindand reversible For a satisfactory performance thetwo threshold values have to be carefully searched fordifferent cover images
(ii) Nonblind watermarking scheme in [13] a nonblindscheme based on wavelet-domain statistical quantityhistogram shifting and clustering (WSQH-SC) isproposed A pixel adjustment is presented at first toavoid the overflow and underflow and a locationmapis used to record the changed pixels This methodachieved good robustness against JPEG JPEG2000and additive Gaussian noise but it is not blind sincethe locations of the changed pixels need to be savedas a part of side information and transmitted to thereceiver side in order to recover the original imageIn [14] the Slantlet transform (SLT) was applied toimage blocks modifying the mean values of the HLand LH subband coefficients to embed the watermarkbits and a second stage of SLT transform is appliedto the LL1 subband embedding another watermarkbit into the HL2 and LH2 subband Because thecoefficients and the mean values are fractional withmore decimal places themean informationwas takenas side information to be sent to the receiver side forthe recovery of the original cover image In order tosolve the nonblind extraction question in [14] theauthors in [15] used IWT on images and randomlyselected 10 coefficients from all the 16 coefficients ina block to compute the amplitude mean of the blockso that the mean information can be embedded intothe image itself for blind extraction
In [16] Coltuc and Chassery proposed a general frame-work for robust reversible watermarking by multiple water-marking First the watermark is embedded into the coverimage with a robust watermarking method and then areversible watermarking method is adopted to embed theinformation (which is used to restore the original coverimage) into the robust watermarked image Suppose 119868 and1198681 are the original image and the robust watermarked imageafter embedding a watermark119908 respectivelyThe embeddingdistortion 119889 = 119868 minus 1198681 is compressed and embedded into therobust watermarked image with the reversible watermarkingmethod At the receiver side if there are no attacks the robustwatermarked image 1198681 and the difference 119889 can be extractedsince the embedding process is reversible Then the originalimage 119868 can be recovered by 119868 = 119889 + 1198681 Furthermorethe watermark can be extracted If the watermarked imagegoes through a JPEG compression operation the robustwatermark can still be extracted This framework is veryinstructive and achieves higher payload and good robustnessagainst JPEG compression
In [17] a robust reversible audio method based on spreadspectrum and amplitude expansion is proposed A robust
payload is embedded at first using the direct-sequencespread-spectrummodulation with the sequence determinedfrom the amplitude expansion in time and frequency of inte-ger modified discrete cosine transform (MDCT) coefficientsThen a reversible payload is embedded into the apertures inthe amplitude histogram that result from amplitude expan-sion of the integer MDCT coefficients to recover the hostaudio This method achieves robustness against some signalprocessing like MP3 compression and additive noise and ifthe watermarked audio remains intact the host audio can berecovered perfectly
In this paper we propose a novel robust and reversibleaudio watermarking scheme based on statistic feature andhistogram shifting in time domain By shifting the histogramof the statistic features in time domain the proposed algo-rithm achieves good robustness and reversibility at the sametime
The rest of the paper is organized as follows The foun-dation work is introduced in Section 2 The proposed water-marking algorithm is described in Section 3 Experimentalresults are presented in Section 4 Section 5 concludes thispaper
2 Algorithmrsquos Principle
This section will introduce the foundation works of the pro-posed robust reversible digital audio watermarking schemeFirstly a robust statistic feature of time domain is introducedthen how to modify the statistic feature to embed thewatermark bit is briefly described
21 Robust Statistic Feature Consider a time-discrete digitalaudio signal 119883 the host signal is first divided into nonover-lapped equal-sized frames We take119873 samples per frame forexample119873 samples as a frame and three samples as a groupare shown in Figure 1 For a sample group (119909119897 119909119898 and 119909119903)the prediction value of the middle sample 119909119898 is calculated byusing two immediate samples as
119909119898 = lceil119909119897 + 1199091199032 rceil (1)
where lceil119909rceil means rounding the elements of 119909 to the nearestinteger towards infinity The prediction error of 119909119898 is
119890119898 = 119909119898 minus 119909119898 (2)
Since the samples in a group are often highly correlatedthe prediction error 119890119898 is expected to be very close to zeroFor a frame with 119873 samples 1198733 prediction errors can becomputed The sum of all the prediction errors in a framedenoted by 119864 is called the statistic feature in this paper Thestatistical feature of a frame is calculated as
119864 =1198733
sum119894=1
119890119894 (3)
where 119890119894 is the prediction error of the 119894th group in the frameThe basic idea of the proposed algorithm is based on thisstatistic property
Advances in Multimedia 3
x1 x2 x3 x4 x5 x6 middot middot middot XNminus2 XNminus1 xN
Group 1 Group 2 middot middot middot Group i
Figure 1 The use of 119873 samples as a frame and three samples as agroup
3002001000 400 500minus200minus300 minus100minus4000
2
4
6
8
10
12
14
Figure 2 The distribution of the 119864 values for the clip track 1
22 Watermark Statistic Feature For each frame one water-mark bit is embedded by shifting the value of the statisticfeature The shifting operation is done by modifying thesamples in a frame Taking track 1 (which is downloadedfrom the website [18]) as example clip Figure 2 shows thedistribution of 119864 values by using 300 samples as a frame andthree samples as a group The rule to modify the statisticvalue is referred to histogram shifting method At first wescan all frames and find out the maximum of the absolute119864 values denoted by 119864max Then a threshold 119879 is set to apositive integer bigger than 119864max As a result all 119864 values arewithin the range [minus119879 119879] For example from Figure 2 we canget that 119864max is 446 so threshold 119879 can be an integer suchas 500 The watermarking rule is to keep the statistic featurewithin [minus119879 119879] if the watermark bit is ldquo0rdquo while the statisticfeature is shifted away from zero by a shifting quantity 119879 + 119866if the watermark bit is ldquo1rdquo To achieve stronger robustnessparameter 119866 is a threshold which is usually set bigger than119879 To reduce the embedding distortion if the embeddedwatermark bit is ldquo1rdquo and the original statistic feature belongsto [0 119879) the statistic feature is shifted to the region [119879 +119866 2119879 + 119866] if the embedded watermark bit is ldquo1rdquo and theoriginal statistic feature belongs to (minus119879 0) the statistic featureis shifted to the region [minus2119879 minus 119866 minus119879 minus 119866] In such a waythe bit-0 region and the bit-1 region are separated by therobust regions (119879 119879 + 119866) and (minus119879 minus 119866 minus119879) For exampleFigure 3 shows the distribution of 119864 values after embeddingwatermark by using clip track 1
The modifying rules are as followsIf the embedded bit is ldquo0rdquo keep the frame unchanged If
the embedded bit is ldquo1rdquo the samples in the frame aremodifiedby
2T
Bit-1region
Robustregion
Bit-0region
Robustregion
Bit-1region
GG
0
5
10
15
20
25
30
35
200010000 3000 4000minus2000minus3000 minus1000minus4000
Figure 3The distribution of the 119864 values of track 1 after embeddingwatermark
119910119896119894 =
119909119896119894 + 120573 if 0 le 119864119896 lt 119879 mod (119894 3) = 2119909119896119894 minus 120573 if minus 119879 lt 119864119896 lt 0 mod (119894 3) = 2119909119896119894 otherwise
(4)
where 119909119896119894 is the 119894th sample in the 119896th frame The index 119894 isin [1 119878] and 119878 is the number of the samples in a frame Theinteger value 120573 is the shifted quantity of a sample
120573 = lceil119879 + 1198661198783 rceil (5)
At the receiver side if the watermarked audio remainsintact the watermark bits can be extracted by
119908119896 = 0 if 119864119896 isin [minus119879 119879]1 otherwise (6)
where 119908119896 is the to be hidden 119896th bit The original audio canbe recovered as
119909119896119894
=
119910119896119894 minus 120573 if119864119896 isin [119879 + 119866 2119879 + 119866] mod (119894 3) = 2119910119896119894 + 120573 if119864119896 isin [minus2119879 minus 119866 minus119879 minus 119866] mod (119894 3) = 2119910119896119894 otherwise
(7)
23 Prevention of OverflowUnderflow For a 16-bit digitalaudio the permission range of the sample value is [minus215 215]Watermark embedding will modify the sample values withthe value 120573 so the overflow or underflow does not occur ifthe original sample values belong to [minus215 + 120573 215 minus 120573] Infact as the value 120573 is very small the original sample values ofmost normal audio belong to [minus215 + 120573 215 minus 120573] Thereforein the proposedmethod there is no overflow or underflow inmost cases If the audio cannot meet this condition we canrecord the location and modify the sample value to the range[minus215 + 120573 215 minus 120573] then the location can be saved as sideinformation and embedded into the audio
4 Advances in Multimedia
Input audio X andwatermark w
Divide audio into framessized S samples
Calculate the statisticfeatures value E of the
frames
Set threshold T(T gt max(abs(E))) and
threshold G (usually G gt T)
wk = 1
No change
Combine the frames to obtainthe watermarked audio
End
Ek isin [0 T]
Yes
Yes
Some samples in the frameare added by value 훽 and
E㰀k = Ek + T + G
No
No
Some samples in the frameare subtracted by value 훽 and
E㰀k = Ek minus T minus G
Figure 4 Watermark embedding process
3 Proposed Algorithm
The embedding and extraction processes are presented indetail as follows
31 Watermark Embedding Figure 4 shows the proposedwatermark embedding process The watermark is embeddedwith the following five steps
Step 1 Divide the original audio 119883 into nonoverlappingframes sized 119878 samples
Step 2 Calculate the statistic features of the frames (119864 values)by referring to (1)ndash(3)
Step 3 Set the threshold values 119879 and 119866 (119879 gt 119864max andusually 119866 gt 119879)Step 4 If the watermark bit is ldquo0rdquo nothing is changed Ifthe bit is ldquo1rdquo shift the statistic feature value with a shiftingquantity 119879 +119866 to embed the watermark bit by modifying thesamples in the frame with value 120573 referring to (4)Step 5 Combine the frames to get the watermarked audio
32 Watermark Extraction If the watermarked audio goesthrough some attacks (such as MP3 compression additivenoise resampling or requantization) the watermark can
still be detected To improve the accuracy of the watermarkextraction three extraction methods and a majority votingsystem are adopted to identify the extracted watermark bycomputing the distorted statistical feature values 1198641015840(i) Extraction 1 Redefine the bit-0 region as [minus119879 minus 1198662 119879 +1198662] and the watermark extraction as
11990810158401 (119894) = 0 if1198641015840 isin [minus119879 minus 1198662 119879 + 1198662]1 otherwise (8)
(ii) Extraction 2 Redefine the bit-0 region as [minus119879 minus 1198663 119879 +1198663] and the watermark extraction as
11990810158402 (119894) = 0 if1198641015840 isin [minus119879 minus 1198663 119879 + 1198663]1 otherwise (9)
(iii) Extraction 3 119896-means clustering algorithm is introducedto extract bits Figure 5 shows the distribution of the 119864 valuesafter MP3 compression and the watermark can be extractedby
11990810158403 (119894) = 0 if1198641015840 isin class 21 if1198641015840 isin class 1 and class 3 (10)
Advances in Multimedia 5
Class 1 Class 3
Class 2
3000
2000
10000
4000
5000
minus20
00
minus30
00
minus40
00
minus10
00
minus50
00
02468
101214161820
Figure 5 The distribution of the statistic feature values of track 1after MP3 compression at 64 kbps
The majority voting system works as
1199081015840 (119894) =
11990810158401 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158401 (119894) = 11990810158403 (119894)11990810158401 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158401 (119894) = 11990810158403 (119894)11990810158402 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158402 (119894) = 11990810158403 (119894)11990810158403 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158401 (119894) = 11990810158403 (119894)
(11)
Eventually three extraction methods and a majorityvoting system are adopted to extract watermark Figure 6shows the proposed watermark extracting process If thewatermarked audio remains intact the watermark can beextracted correctly and the original audio can be recoveredas the following steps
Step 1 Divide the watermarked audio 119883119908 into nonoverlap-ping frames sized 119878 samples
Step 2 Calculate the statistic feature of the frames (1198641015840 values)by referring to (1)ndash(3)
Step 3 Extract the watermark with three extraction methodsand identify the watermark with the majority voting systemby referring to (8)ndash(11)
Step 4 The original audio can be recovered by modifying thesamples in the frame with value 120573 referring to (7)Step 5 Combine the frames to get the original audio
If the watermarked audio goes through some attacks theoriginal audio cannot be recovered exactly so we focus onthe watermark extraction and the watermark is extracted asfollows
Step 1 Divide the watermarked audio 119883119908 into nonoverlap-ping frames sized 119878 samples
Step 2 Calculate the statistic feature values of the frames 1198641015840by referring to (1)ndash(3)
Step 3 Extract the watermark with three extraction methodsand identify the watermark with the majority voting systemby referring to (8)ndash(11)
4 Experimental Results
In this section 7WAVaudio file of the sample rate of 441 KHzand 16 bits per sample (tracks 1 2 3 4 5 6 and 7 [18]) are usedas example clips to evaluate the performance of the proposedalgorithm The payload of our method only depends on thelength of a frame 119878 for a time-discrete digital audio signal119883in length119872 the pure payload can be calculated by
119862 = lfloor119872119878 rfloor (12)
In the experiment the watermark is a pseudo-randomsequence in length of 1000 bits The imperceptibility is firstanalyzed by the SNR standard at different threshold valuesand different sample numbers per frame Then robustnesstesting against MP3 compression additive noise (AWGN)resampling (441-16-441 kHz) and requantization (16-8-16bits) are reported by using the software CoolEditPro v21
41 Imperceptibility Test The imperceptibility is measuredby the embedding distortion In the proposed scheme thedistortion is caused by the shifting quantity on the samplesdepending on thresholds 119879 119866 and the length of a frame 119878Since 119879 is set at first we only investigate the influence of 119866and 119878 on SNR
Figure 7 plots the relationship between SNR and thethreshold 119866 for different clips at the same threshold 119879 and119878 From this figure we can conclude that with the increase of119866 the SNR value drops The reason is that the larger 119866 isthe larger shifting quantity is used so the larger embeddingdistortion is caused As a result SNR value drops
Figure 8 plots the relationship between SNR and thelength of a frame 119878 for different clips at the same thresholds119879 and 119866 We can see from this figure that the larger 119878is the higher SNR value is achieved The reason is thatwith the increase of 119878 the shifting quantity for every singlesample drops so the SNR values rise due to the fact that theembedding distortion is reduced Consequently the framelength 119878 will influence Maximum Embedding Capacity andSNR value directly the Maximum Embedding Capacity ishigher when 119878 is smaller according to (12) and the SNR valueis higher when 119878 is larger according to Figure 8 To considerthe balance between Maximum Embedding Capacity andSNR value we have found that the value of 119878within the rangeof 300 to 600 can usually achieve a satisfactory result after aset of experiments
42 Robustness Testing To test the robustness of the pro-posed scheme a set of experiments has been taken ontracks 1ndash7 Table 1 shows the results in which RP meansresampling (441-16-441 kHz) operations while RQ meansrequantization (16-8-16 bits) operations We can observefrom this table that all the example clips can achieve therobustness against MP3 compression at 64Kbps For track1 the watermark bits can be correctly extracted under the
6 Advances in Multimedia
Input audio Xw
Divide audio into framessized S samples
Calculate the statisticfeatures value E㰀 of the
frames
Three extraction methodsand a majority voting system
Watermark w
End
Yes
Yes
No
No
E㰀k isin [T + G 2T + G]
No change
E㰀k isin [minus2T minus G minusT minus G]
Combine the frames toobtain the original audio
Some samples in the frameare subtracted by value 훽 and
Ek = E㰀k minus T minus G
Some samples in the frameare added by value 훽 and
Ek = E㰀k + T + G
Figure 6 Watermark extracting process
3500 4000 4500 5000 5500 6000 6500 7000 75003000Threshold G
40
45
50
55
60
SNR
(dB)
Track 1 T = 500 S = 300Track 2 T = 5300 S = 300Track 3 T = 300 S = 300Track 4 T = 9900 S = 300
Track 5 T = 200 S = 300Track 6 T = 200 S = 300Track 7 T = 1700 S = 300
Figure 7 Relationship between SNR and threshold 119866
MP3 compression of 48KbpsThe robustness against additivenoise is also satisfactory Even with the noise intensity at25 dB the BER (bit error rate) values are less than 10except for track 1 Besides the watermark robustness againstresampling and requantization operations is perfect and thehidden bits can be recovered without errors
350 400 450 500 550 600300Length S
1015202530354045505560
SNR
(dB)
Track 1 T = 550G = 3000Track 2 T = 6200G = 7000Track 3 T = 1700G = 4000Track 4 T = 16000G = 17000
Track 5 T = 300G = 3000Track 6 T = 200G = 3000Track 7 T = 2700G = 3000
Figure 8 Relationship between SNR and the frame length 119878
As shown in Figure 3 the robustness of the proposedmethod is originated from the robust region The robustregion depends on threshold 119866 The larger 119866 is the largerthe robust region is and the stronger robustness is Figure 9
Advances in Multimedia 7
Table 1 Performance of the proposed method
Audio 119878 119879 119866 SNR MP3 (Kbps) AWGN (35 dB) AWGN (30 dB) AWGN (25 dB) RP RQTrack 1 420 500 3000 5474 48 21000 771000 2511000 01000 01000Track 2 600 6200 7000 4798 64 01000 01000 01000 01000 01000Track 3 510 600 3000 5409 48 01000 01000 441000 01000 01000Track 4 300 9900 10000 3899 64 01000 01000 01000 01000 01000Track 5 300 200 3000 5618 48 01000 251000 771000 01000 01000Track 6 600 200 3000 495 48 01000 01000 461000 01000 01000Track 7 300 1600 4000 4378 48 01000 01000 01000 01000 01000
Table 2 Performance of the proposed method with different length of a frame 119878 on track 1
Track S T G MP3 AWGN AWGN AWGN(Kbps) (35 dB) (30 dB) (25 dB)
Track 1 300 500 3000 80 91000 901000 2991000Track 1 420 500 3000 56 121000 1711000 3201000Track 1 510 550 3000 80 281000 1711000 1581000Track 1 600 550 3000 48 281000 2211000 3901000Track 6 300 200 3000 48 01000 01000 31000Track 6 420 200 3000 56 281000 01000 231000Track 7 300 200 4000 48 01000 01000 01000Track 7 420 200 4000 56 01000 01000 01000
Track 1 G = 5000Track 1 G = 6000Track 2 G = 6000Track 2 G = 7000
Track 3 G = 3000Track 3 G = 4000Track 4 G = 9000Track 4 G = 10000
50 55 60 65 70 75 80 85 90 95 10045MP3 compression bit rate
0
001
002
003
004
005
Bit e
rror
rate
(BER
)
Figure 9 Relationship between MP3 bit rate and threshold 119866
supports the conclusion Figure 9 shows the bit error rate(BER) at different threshold 119866 for the same audio withsame threshold 119879 The lower BER means that the strongerrobustness is achieved We can find that as threshold 119866increases the BER drops and the robustness rises
Take track 1 as example clip Figure 10 shows the bit errorrate of the extracted watermark with different thresholds 119866against additive noises with the same 119879 and 119878 (119879 = 500 119878 =300) We can see that the larger 119866 is the smaller bit errorrate is and the better robustness is As threshold 119866 increasesthe robustness becomes stronger In the application we canadjust the parameter 119866 to achieve ideal robustness On the
25 30 35 4020AWGN (dB)
0005
01015
02025
03035
04045
05
Bit e
rror
rate
(BER
)
Track 1 G = 3000Track 1 G = 4000Track 1 G = 5000
Track 1 G = 6000Track 1 G = 7500
Figure 10 The BER values at different AWGN with differentthreshold 119866
other hand with the increase of 119866 the SNR value drops Toconsider the balance between SNR value and robustness wehave found that the value of119866within 3000 to 5000 can usuallyachieve a satisfactory result after a set of experiments
To evaluate the effect of the frame length 119878 on therobustness performance a set of experiments has been takenon track 1 track 6 and track 7 Table 2 lists the results Wecan observe that for the same audio with the same 119879 and 119866such as 119879 = 500 and 119866 = 3000 for track 1 as 119878 increasesthe robustness against MP3 compression will strengthen butfor track 6 and track 7 as 119878 increases the robustness against
8 Advances in Multimedia
Table3Com
paris
onwith
metho
din
[17]
Track
Metho
din
[17]
Prop
osed
metho
dPayload
ODG
MP3
AWGN
119878119879
119866Payload
ODG
MP3
MP3
AWGN
(bits)
(128
Kbps)
(36d
B)(bits)
(80K
bps)
(64K
bps)
(36d
B)Track32
216
minus245
00
150
1500
3000
1000
minus137
002
0Track35
216
minus131
012
150
3500
5000
1000
minus158
008
0Track65
216
minus121
06
150
302000
1000
minus023
001
12
Track66
216
minus024
06
9660
0060
001000
minus105
03
12
0Track69
216
minus031
14
300
400
2000
1000
minus023
00
0
Advances in Multimedia 9
MP3 compression drops so the effect of the frame length 119878on the robustness against MP3 compression is unstable Theinfluence on AWGN is little
For fair comparison with the method in [17] we use thesame host signals (tracks 32 35 65 66 and 69) downloadedfrom sound quality assessment material (SQAM) collection[19] Table 3 shows the robustness testing results against MP3compression and additive noise (AWGN) operations We canobserve that the method in [17] can carry 216 bits and resisttheMP3 compression at 128Kbpswhile the proposedmethodcan resist the MP3 compression at 64Kbps with 1000 bitsembedded In addition in our method the BER under theAWGN of 35 dB is less than that of the method in [17]In other words the proposed method can provide largerembedding capacity and obtain stronger robustness againstMP3 compression and AWGN attacks The imperceptibilityis evaluated by using the ODG standardThe closer the ODGvalue to 0 the better the imperceptibility For the table it isnoted that the imperceptibility of the proposed method isbetter except for the clips track 35 and track 66 The reason isthat119864max values of track 35 and track 66 are bigger As a resultthresholds 119879 and 119866 are also larger and more embeddingdistortion will be caused
5 Conclusions
In this paper we proposed a robust and reversible audiowatermarking method by shifting the histogram of thestatistical feature values in time domainThe statistical featureis calculated as the sum of the prediction errors in a frameSince the audio clip has a larger number of samples andeach frame can hold enough elements the statistical featureis robust to those common signal processing operationsConsidering that the distribution of the statistical featurevalues may be distorted to some extent three extractionmethods and the majority voting system are designed forthe watermark detection Experimental results have shownthat thousands of bits can be reversibly embedded and thewatermark bits can resist MP3 compression of 64 kbps andadditive noise of 35 dB Comparingwith the existing excellentmethod the proposed method can embed more watermarkbits and achieve stronger robustness
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This work was supported by NSFC (no 61272414) and OpenResearch Fund from the State Key Laboratory of InformationSecurity (no 2016-MS-07)
References
[1] Y Xiang I Natgunanathan S GuoW Zhou and S NahavandildquoPatchwork-based audio watermarking method robust to de-synchronization attacksrdquo IEEE Transactions on Audio Speechand Language Processing vol 22 no 9 pp 1413ndash1423 2014
[2] J Fridrich M Goljan and R Du ldquoLossless data embeddingmdashnew paradigm in digital watermarkingrdquo EURASIP Journal onAdvances in Signal Processing vol 2002 no 2 pp 185ndash196 2002
[3] Y Q Shi Z Ni D Zou C Liang and G Xuan ldquoLosslessdata hiding fundamentals algorithms and applicationsrdquo inProceeding of the IEEE International Symposium on Circuits andSystems vol 2 pp 313ndash336 2004
[4] S Lee CD Yoo andTKalker ldquoReversible imagewatermarkingbased on integer-to-integer wavelet transformrdquo IEEE Transac-tions on Information Forensics and Security vol 2 no 3 pp 321ndash330 2010
[5] X Li B Yang and T Zeng ldquoEfficient reversible watermarkingbased on adaptive prediction-error expansion and pixel selec-tionrdquo IEEE Transactions on Image Processing vol 20 no 12 pp3524ndash3533 2000
[6] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hidingrdquo in Proceeding of the IEEE Interna-tional Conference onMultimedia and Expo (ICME rsquo2004) vol 3pp 2199ndash2202 Taipei Taiwan June 2004
[7] C De Vleeschouwer J Delaigle and B Macq ldquoCircularinterpretation of histogram for reversible watermarkingrdquo inProceeding of the IEEE 4th Workshop on Multimedia SignalProcessing pp 345ndash350 Cannes France 2001
[8] C De Vleeschouwer J F Delaigle and B Macq ldquoCircularinterpretation of bijective transformations in lossless water-marking for media asset managementrdquo IEEE Transactions onMultimedia vol 5 no 1 pp 97ndash105 2003
[9] D Zou YQ Shi ZNi andW Su ldquoA semi-fragile lossless digitalwatermarking scheme based on integer wavelet transformrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 10 pp 1294ndash1300 2006
[10] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hiding designed for semi-fragile imageauthenticationrdquo IEEE Transactions on Circuits and Systems forVideo Technology vol 18 no 4 pp 890ndash896 2008
[11] L An X Gao C Deng and F Ji ldquoRobust lossless data hidingAnalysis and evaluationrdquo in Procceding of the InternationalConference on High Performance Computing and Simulation(HPCS rsquo10) pp 512ndash516 July 2010
[12] X-T Zeng L-D Ping and X-Z Pan ldquoA lossless robust datahiding schemerdquo Pattern Recognition vol 43 no 4 pp 1656ndash1667 2010
[13] L An X Gao X Li D Tao C Deng and J Li ldquoRobustreversible watermarking via clustering and enhanced pixel-wisemaskingrdquo IEEE Transactions on Image Processing vol 21 no 8pp 3598ndash3611 2012
[14] R Thabit and B E Khoo ldquoCapacity improved robust losslessimage watermarkingrdquo IET Image Processing vol 8 no 11 pp662ndash670 2014
[15] S Xiang and Y Wang ldquoDistortion-free robust reversible water-marking by modifying and recording iwt means of imageblocksrdquo in Proceeding of the 14th International Workshop(IWDW rsquo15) Tokyo Japan October 2015
[16] D Coltuc and J Chassery ldquoDistortion-free robust watermark-ing a case studyrdquo in Security Steganography andWatermarkingof Multimedia Contents vol 6505 of Proceedings of SPIE pp588ndash595 San Jose Calif USA 2007
[17] A Nishimura ldquoReversible and robust audio watermarkingbased on spread spectrum and amplitude expansionrdquo in Inter-national Workshop on Digital Watermarking (IWDW rsquo14) vol9023 of Lecture Notes in Computer Science pp 215ndash229
10 Advances in Multimedia
[18] Massachusetts Institute of Technology (MIT) Audio Databasehttpsoundmediamitedumediaphp
[19] EBUCommittee Sound quality assessmentmaterial recordingsfor subjective tests httpstechebuchpublicationssqamcd
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal of
Volume 201
Submit your manuscripts athttpswwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 201
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
Advances in Multimedia 3
x1 x2 x3 x4 x5 x6 middot middot middot XNminus2 XNminus1 xN
Group 1 Group 2 middot middot middot Group i
Figure 1 The use of 119873 samples as a frame and three samples as agroup
3002001000 400 500minus200minus300 minus100minus4000
2
4
6
8
10
12
14
Figure 2 The distribution of the 119864 values for the clip track 1
22 Watermark Statistic Feature For each frame one water-mark bit is embedded by shifting the value of the statisticfeature The shifting operation is done by modifying thesamples in a frame Taking track 1 (which is downloadedfrom the website [18]) as example clip Figure 2 shows thedistribution of 119864 values by using 300 samples as a frame andthree samples as a group The rule to modify the statisticvalue is referred to histogram shifting method At first wescan all frames and find out the maximum of the absolute119864 values denoted by 119864max Then a threshold 119879 is set to apositive integer bigger than 119864max As a result all 119864 values arewithin the range [minus119879 119879] For example from Figure 2 we canget that 119864max is 446 so threshold 119879 can be an integer suchas 500 The watermarking rule is to keep the statistic featurewithin [minus119879 119879] if the watermark bit is ldquo0rdquo while the statisticfeature is shifted away from zero by a shifting quantity 119879 + 119866if the watermark bit is ldquo1rdquo To achieve stronger robustnessparameter 119866 is a threshold which is usually set bigger than119879 To reduce the embedding distortion if the embeddedwatermark bit is ldquo1rdquo and the original statistic feature belongsto [0 119879) the statistic feature is shifted to the region [119879 +119866 2119879 + 119866] if the embedded watermark bit is ldquo1rdquo and theoriginal statistic feature belongs to (minus119879 0) the statistic featureis shifted to the region [minus2119879 minus 119866 minus119879 minus 119866] In such a waythe bit-0 region and the bit-1 region are separated by therobust regions (119879 119879 + 119866) and (minus119879 minus 119866 minus119879) For exampleFigure 3 shows the distribution of 119864 values after embeddingwatermark by using clip track 1
The modifying rules are as followsIf the embedded bit is ldquo0rdquo keep the frame unchanged If
the embedded bit is ldquo1rdquo the samples in the frame aremodifiedby
2T
Bit-1region
Robustregion
Bit-0region
Robustregion
Bit-1region
GG
0
5
10
15
20
25
30
35
200010000 3000 4000minus2000minus3000 minus1000minus4000
Figure 3The distribution of the 119864 values of track 1 after embeddingwatermark
119910119896119894 =
119909119896119894 + 120573 if 0 le 119864119896 lt 119879 mod (119894 3) = 2119909119896119894 minus 120573 if minus 119879 lt 119864119896 lt 0 mod (119894 3) = 2119909119896119894 otherwise
(4)
where 119909119896119894 is the 119894th sample in the 119896th frame The index 119894 isin [1 119878] and 119878 is the number of the samples in a frame Theinteger value 120573 is the shifted quantity of a sample
120573 = lceil119879 + 1198661198783 rceil (5)
At the receiver side if the watermarked audio remainsintact the watermark bits can be extracted by
119908119896 = 0 if 119864119896 isin [minus119879 119879]1 otherwise (6)
where 119908119896 is the to be hidden 119896th bit The original audio canbe recovered as
119909119896119894
=
119910119896119894 minus 120573 if119864119896 isin [119879 + 119866 2119879 + 119866] mod (119894 3) = 2119910119896119894 + 120573 if119864119896 isin [minus2119879 minus 119866 minus119879 minus 119866] mod (119894 3) = 2119910119896119894 otherwise
(7)
23 Prevention of OverflowUnderflow For a 16-bit digitalaudio the permission range of the sample value is [minus215 215]Watermark embedding will modify the sample values withthe value 120573 so the overflow or underflow does not occur ifthe original sample values belong to [minus215 + 120573 215 minus 120573] Infact as the value 120573 is very small the original sample values ofmost normal audio belong to [minus215 + 120573 215 minus 120573] Thereforein the proposedmethod there is no overflow or underflow inmost cases If the audio cannot meet this condition we canrecord the location and modify the sample value to the range[minus215 + 120573 215 minus 120573] then the location can be saved as sideinformation and embedded into the audio
4 Advances in Multimedia
Input audio X andwatermark w
Divide audio into framessized S samples
Calculate the statisticfeatures value E of the
frames
Set threshold T(T gt max(abs(E))) and
threshold G (usually G gt T)
wk = 1
No change
Combine the frames to obtainthe watermarked audio
End
Ek isin [0 T]
Yes
Yes
Some samples in the frameare added by value 훽 and
E㰀k = Ek + T + G
No
No
Some samples in the frameare subtracted by value 훽 and
E㰀k = Ek minus T minus G
Figure 4 Watermark embedding process
3 Proposed Algorithm
The embedding and extraction processes are presented indetail as follows
31 Watermark Embedding Figure 4 shows the proposedwatermark embedding process The watermark is embeddedwith the following five steps
Step 1 Divide the original audio 119883 into nonoverlappingframes sized 119878 samples
Step 2 Calculate the statistic features of the frames (119864 values)by referring to (1)ndash(3)
Step 3 Set the threshold values 119879 and 119866 (119879 gt 119864max andusually 119866 gt 119879)Step 4 If the watermark bit is ldquo0rdquo nothing is changed Ifthe bit is ldquo1rdquo shift the statistic feature value with a shiftingquantity 119879 +119866 to embed the watermark bit by modifying thesamples in the frame with value 120573 referring to (4)Step 5 Combine the frames to get the watermarked audio
32 Watermark Extraction If the watermarked audio goesthrough some attacks (such as MP3 compression additivenoise resampling or requantization) the watermark can
still be detected To improve the accuracy of the watermarkextraction three extraction methods and a majority votingsystem are adopted to identify the extracted watermark bycomputing the distorted statistical feature values 1198641015840(i) Extraction 1 Redefine the bit-0 region as [minus119879 minus 1198662 119879 +1198662] and the watermark extraction as
11990810158401 (119894) = 0 if1198641015840 isin [minus119879 minus 1198662 119879 + 1198662]1 otherwise (8)
(ii) Extraction 2 Redefine the bit-0 region as [minus119879 minus 1198663 119879 +1198663] and the watermark extraction as
11990810158402 (119894) = 0 if1198641015840 isin [minus119879 minus 1198663 119879 + 1198663]1 otherwise (9)
(iii) Extraction 3 119896-means clustering algorithm is introducedto extract bits Figure 5 shows the distribution of the 119864 valuesafter MP3 compression and the watermark can be extractedby
11990810158403 (119894) = 0 if1198641015840 isin class 21 if1198641015840 isin class 1 and class 3 (10)
Advances in Multimedia 5
Class 1 Class 3
Class 2
3000
2000
10000
4000
5000
minus20
00
minus30
00
minus40
00
minus10
00
minus50
00
02468
101214161820
Figure 5 The distribution of the statistic feature values of track 1after MP3 compression at 64 kbps
The majority voting system works as
1199081015840 (119894) =
11990810158401 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158401 (119894) = 11990810158403 (119894)11990810158401 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158401 (119894) = 11990810158403 (119894)11990810158402 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158402 (119894) = 11990810158403 (119894)11990810158403 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158401 (119894) = 11990810158403 (119894)
(11)
Eventually three extraction methods and a majorityvoting system are adopted to extract watermark Figure 6shows the proposed watermark extracting process If thewatermarked audio remains intact the watermark can beextracted correctly and the original audio can be recoveredas the following steps
Step 1 Divide the watermarked audio 119883119908 into nonoverlap-ping frames sized 119878 samples
Step 2 Calculate the statistic feature of the frames (1198641015840 values)by referring to (1)ndash(3)
Step 3 Extract the watermark with three extraction methodsand identify the watermark with the majority voting systemby referring to (8)ndash(11)
Step 4 The original audio can be recovered by modifying thesamples in the frame with value 120573 referring to (7)Step 5 Combine the frames to get the original audio
If the watermarked audio goes through some attacks theoriginal audio cannot be recovered exactly so we focus onthe watermark extraction and the watermark is extracted asfollows
Step 1 Divide the watermarked audio 119883119908 into nonoverlap-ping frames sized 119878 samples
Step 2 Calculate the statistic feature values of the frames 1198641015840by referring to (1)ndash(3)
Step 3 Extract the watermark with three extraction methodsand identify the watermark with the majority voting systemby referring to (8)ndash(11)
4 Experimental Results
In this section 7WAVaudio file of the sample rate of 441 KHzand 16 bits per sample (tracks 1 2 3 4 5 6 and 7 [18]) are usedas example clips to evaluate the performance of the proposedalgorithm The payload of our method only depends on thelength of a frame 119878 for a time-discrete digital audio signal119883in length119872 the pure payload can be calculated by
119862 = lfloor119872119878 rfloor (12)
In the experiment the watermark is a pseudo-randomsequence in length of 1000 bits The imperceptibility is firstanalyzed by the SNR standard at different threshold valuesand different sample numbers per frame Then robustnesstesting against MP3 compression additive noise (AWGN)resampling (441-16-441 kHz) and requantization (16-8-16bits) are reported by using the software CoolEditPro v21
41 Imperceptibility Test The imperceptibility is measuredby the embedding distortion In the proposed scheme thedistortion is caused by the shifting quantity on the samplesdepending on thresholds 119879 119866 and the length of a frame 119878Since 119879 is set at first we only investigate the influence of 119866and 119878 on SNR
Figure 7 plots the relationship between SNR and thethreshold 119866 for different clips at the same threshold 119879 and119878 From this figure we can conclude that with the increase of119866 the SNR value drops The reason is that the larger 119866 isthe larger shifting quantity is used so the larger embeddingdistortion is caused As a result SNR value drops
Figure 8 plots the relationship between SNR and thelength of a frame 119878 for different clips at the same thresholds119879 and 119866 We can see from this figure that the larger 119878is the higher SNR value is achieved The reason is thatwith the increase of 119878 the shifting quantity for every singlesample drops so the SNR values rise due to the fact that theembedding distortion is reduced Consequently the framelength 119878 will influence Maximum Embedding Capacity andSNR value directly the Maximum Embedding Capacity ishigher when 119878 is smaller according to (12) and the SNR valueis higher when 119878 is larger according to Figure 8 To considerthe balance between Maximum Embedding Capacity andSNR value we have found that the value of 119878within the rangeof 300 to 600 can usually achieve a satisfactory result after aset of experiments
42 Robustness Testing To test the robustness of the pro-posed scheme a set of experiments has been taken ontracks 1ndash7 Table 1 shows the results in which RP meansresampling (441-16-441 kHz) operations while RQ meansrequantization (16-8-16 bits) operations We can observefrom this table that all the example clips can achieve therobustness against MP3 compression at 64Kbps For track1 the watermark bits can be correctly extracted under the
6 Advances in Multimedia
Input audio Xw
Divide audio into framessized S samples
Calculate the statisticfeatures value E㰀 of the
frames
Three extraction methodsand a majority voting system
Watermark w
End
Yes
Yes
No
No
E㰀k isin [T + G 2T + G]
No change
E㰀k isin [minus2T minus G minusT minus G]
Combine the frames toobtain the original audio
Some samples in the frameare subtracted by value 훽 and
Ek = E㰀k minus T minus G
Some samples in the frameare added by value 훽 and
Ek = E㰀k + T + G
Figure 6 Watermark extracting process
3500 4000 4500 5000 5500 6000 6500 7000 75003000Threshold G
40
45
50
55
60
SNR
(dB)
Track 1 T = 500 S = 300Track 2 T = 5300 S = 300Track 3 T = 300 S = 300Track 4 T = 9900 S = 300
Track 5 T = 200 S = 300Track 6 T = 200 S = 300Track 7 T = 1700 S = 300
Figure 7 Relationship between SNR and threshold 119866
MP3 compression of 48KbpsThe robustness against additivenoise is also satisfactory Even with the noise intensity at25 dB the BER (bit error rate) values are less than 10except for track 1 Besides the watermark robustness againstresampling and requantization operations is perfect and thehidden bits can be recovered without errors
350 400 450 500 550 600300Length S
1015202530354045505560
SNR
(dB)
Track 1 T = 550G = 3000Track 2 T = 6200G = 7000Track 3 T = 1700G = 4000Track 4 T = 16000G = 17000
Track 5 T = 300G = 3000Track 6 T = 200G = 3000Track 7 T = 2700G = 3000
Figure 8 Relationship between SNR and the frame length 119878
As shown in Figure 3 the robustness of the proposedmethod is originated from the robust region The robustregion depends on threshold 119866 The larger 119866 is the largerthe robust region is and the stronger robustness is Figure 9
Advances in Multimedia 7
Table 1 Performance of the proposed method
Audio 119878 119879 119866 SNR MP3 (Kbps) AWGN (35 dB) AWGN (30 dB) AWGN (25 dB) RP RQTrack 1 420 500 3000 5474 48 21000 771000 2511000 01000 01000Track 2 600 6200 7000 4798 64 01000 01000 01000 01000 01000Track 3 510 600 3000 5409 48 01000 01000 441000 01000 01000Track 4 300 9900 10000 3899 64 01000 01000 01000 01000 01000Track 5 300 200 3000 5618 48 01000 251000 771000 01000 01000Track 6 600 200 3000 495 48 01000 01000 461000 01000 01000Track 7 300 1600 4000 4378 48 01000 01000 01000 01000 01000
Table 2 Performance of the proposed method with different length of a frame 119878 on track 1
Track S T G MP3 AWGN AWGN AWGN(Kbps) (35 dB) (30 dB) (25 dB)
Track 1 300 500 3000 80 91000 901000 2991000Track 1 420 500 3000 56 121000 1711000 3201000Track 1 510 550 3000 80 281000 1711000 1581000Track 1 600 550 3000 48 281000 2211000 3901000Track 6 300 200 3000 48 01000 01000 31000Track 6 420 200 3000 56 281000 01000 231000Track 7 300 200 4000 48 01000 01000 01000Track 7 420 200 4000 56 01000 01000 01000
Track 1 G = 5000Track 1 G = 6000Track 2 G = 6000Track 2 G = 7000
Track 3 G = 3000Track 3 G = 4000Track 4 G = 9000Track 4 G = 10000
50 55 60 65 70 75 80 85 90 95 10045MP3 compression bit rate
0
001
002
003
004
005
Bit e
rror
rate
(BER
)
Figure 9 Relationship between MP3 bit rate and threshold 119866
supports the conclusion Figure 9 shows the bit error rate(BER) at different threshold 119866 for the same audio withsame threshold 119879 The lower BER means that the strongerrobustness is achieved We can find that as threshold 119866increases the BER drops and the robustness rises
Take track 1 as example clip Figure 10 shows the bit errorrate of the extracted watermark with different thresholds 119866against additive noises with the same 119879 and 119878 (119879 = 500 119878 =300) We can see that the larger 119866 is the smaller bit errorrate is and the better robustness is As threshold 119866 increasesthe robustness becomes stronger In the application we canadjust the parameter 119866 to achieve ideal robustness On the
25 30 35 4020AWGN (dB)
0005
01015
02025
03035
04045
05
Bit e
rror
rate
(BER
)
Track 1 G = 3000Track 1 G = 4000Track 1 G = 5000
Track 1 G = 6000Track 1 G = 7500
Figure 10 The BER values at different AWGN with differentthreshold 119866
other hand with the increase of 119866 the SNR value drops Toconsider the balance between SNR value and robustness wehave found that the value of119866within 3000 to 5000 can usuallyachieve a satisfactory result after a set of experiments
To evaluate the effect of the frame length 119878 on therobustness performance a set of experiments has been takenon track 1 track 6 and track 7 Table 2 lists the results Wecan observe that for the same audio with the same 119879 and 119866such as 119879 = 500 and 119866 = 3000 for track 1 as 119878 increasesthe robustness against MP3 compression will strengthen butfor track 6 and track 7 as 119878 increases the robustness against
8 Advances in Multimedia
Table3Com
paris
onwith
metho
din
[17]
Track
Metho
din
[17]
Prop
osed
metho
dPayload
ODG
MP3
AWGN
119878119879
119866Payload
ODG
MP3
MP3
AWGN
(bits)
(128
Kbps)
(36d
B)(bits)
(80K
bps)
(64K
bps)
(36d
B)Track32
216
minus245
00
150
1500
3000
1000
minus137
002
0Track35
216
minus131
012
150
3500
5000
1000
minus158
008
0Track65
216
minus121
06
150
302000
1000
minus023
001
12
Track66
216
minus024
06
9660
0060
001000
minus105
03
12
0Track69
216
minus031
14
300
400
2000
1000
minus023
00
0
Advances in Multimedia 9
MP3 compression drops so the effect of the frame length 119878on the robustness against MP3 compression is unstable Theinfluence on AWGN is little
For fair comparison with the method in [17] we use thesame host signals (tracks 32 35 65 66 and 69) downloadedfrom sound quality assessment material (SQAM) collection[19] Table 3 shows the robustness testing results against MP3compression and additive noise (AWGN) operations We canobserve that the method in [17] can carry 216 bits and resisttheMP3 compression at 128Kbpswhile the proposedmethodcan resist the MP3 compression at 64Kbps with 1000 bitsembedded In addition in our method the BER under theAWGN of 35 dB is less than that of the method in [17]In other words the proposed method can provide largerembedding capacity and obtain stronger robustness againstMP3 compression and AWGN attacks The imperceptibilityis evaluated by using the ODG standardThe closer the ODGvalue to 0 the better the imperceptibility For the table it isnoted that the imperceptibility of the proposed method isbetter except for the clips track 35 and track 66 The reason isthat119864max values of track 35 and track 66 are bigger As a resultthresholds 119879 and 119866 are also larger and more embeddingdistortion will be caused
5 Conclusions
In this paper we proposed a robust and reversible audiowatermarking method by shifting the histogram of thestatistical feature values in time domainThe statistical featureis calculated as the sum of the prediction errors in a frameSince the audio clip has a larger number of samples andeach frame can hold enough elements the statistical featureis robust to those common signal processing operationsConsidering that the distribution of the statistical featurevalues may be distorted to some extent three extractionmethods and the majority voting system are designed forthe watermark detection Experimental results have shownthat thousands of bits can be reversibly embedded and thewatermark bits can resist MP3 compression of 64 kbps andadditive noise of 35 dB Comparingwith the existing excellentmethod the proposed method can embed more watermarkbits and achieve stronger robustness
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This work was supported by NSFC (no 61272414) and OpenResearch Fund from the State Key Laboratory of InformationSecurity (no 2016-MS-07)
References
[1] Y Xiang I Natgunanathan S GuoW Zhou and S NahavandildquoPatchwork-based audio watermarking method robust to de-synchronization attacksrdquo IEEE Transactions on Audio Speechand Language Processing vol 22 no 9 pp 1413ndash1423 2014
[2] J Fridrich M Goljan and R Du ldquoLossless data embeddingmdashnew paradigm in digital watermarkingrdquo EURASIP Journal onAdvances in Signal Processing vol 2002 no 2 pp 185ndash196 2002
[3] Y Q Shi Z Ni D Zou C Liang and G Xuan ldquoLosslessdata hiding fundamentals algorithms and applicationsrdquo inProceeding of the IEEE International Symposium on Circuits andSystems vol 2 pp 313ndash336 2004
[4] S Lee CD Yoo andTKalker ldquoReversible imagewatermarkingbased on integer-to-integer wavelet transformrdquo IEEE Transac-tions on Information Forensics and Security vol 2 no 3 pp 321ndash330 2010
[5] X Li B Yang and T Zeng ldquoEfficient reversible watermarkingbased on adaptive prediction-error expansion and pixel selec-tionrdquo IEEE Transactions on Image Processing vol 20 no 12 pp3524ndash3533 2000
[6] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hidingrdquo in Proceeding of the IEEE Interna-tional Conference onMultimedia and Expo (ICME rsquo2004) vol 3pp 2199ndash2202 Taipei Taiwan June 2004
[7] C De Vleeschouwer J Delaigle and B Macq ldquoCircularinterpretation of histogram for reversible watermarkingrdquo inProceeding of the IEEE 4th Workshop on Multimedia SignalProcessing pp 345ndash350 Cannes France 2001
[8] C De Vleeschouwer J F Delaigle and B Macq ldquoCircularinterpretation of bijective transformations in lossless water-marking for media asset managementrdquo IEEE Transactions onMultimedia vol 5 no 1 pp 97ndash105 2003
[9] D Zou YQ Shi ZNi andW Su ldquoA semi-fragile lossless digitalwatermarking scheme based on integer wavelet transformrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 10 pp 1294ndash1300 2006
[10] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hiding designed for semi-fragile imageauthenticationrdquo IEEE Transactions on Circuits and Systems forVideo Technology vol 18 no 4 pp 890ndash896 2008
[11] L An X Gao C Deng and F Ji ldquoRobust lossless data hidingAnalysis and evaluationrdquo in Procceding of the InternationalConference on High Performance Computing and Simulation(HPCS rsquo10) pp 512ndash516 July 2010
[12] X-T Zeng L-D Ping and X-Z Pan ldquoA lossless robust datahiding schemerdquo Pattern Recognition vol 43 no 4 pp 1656ndash1667 2010
[13] L An X Gao X Li D Tao C Deng and J Li ldquoRobustreversible watermarking via clustering and enhanced pixel-wisemaskingrdquo IEEE Transactions on Image Processing vol 21 no 8pp 3598ndash3611 2012
[14] R Thabit and B E Khoo ldquoCapacity improved robust losslessimage watermarkingrdquo IET Image Processing vol 8 no 11 pp662ndash670 2014
[15] S Xiang and Y Wang ldquoDistortion-free robust reversible water-marking by modifying and recording iwt means of imageblocksrdquo in Proceeding of the 14th International Workshop(IWDW rsquo15) Tokyo Japan October 2015
[16] D Coltuc and J Chassery ldquoDistortion-free robust watermark-ing a case studyrdquo in Security Steganography andWatermarkingof Multimedia Contents vol 6505 of Proceedings of SPIE pp588ndash595 San Jose Calif USA 2007
[17] A Nishimura ldquoReversible and robust audio watermarkingbased on spread spectrum and amplitude expansionrdquo in Inter-national Workshop on Digital Watermarking (IWDW rsquo14) vol9023 of Lecture Notes in Computer Science pp 215ndash229
10 Advances in Multimedia
[18] Massachusetts Institute of Technology (MIT) Audio Databasehttpsoundmediamitedumediaphp
[19] EBUCommittee Sound quality assessmentmaterial recordingsfor subjective tests httpstechebuchpublicationssqamcd
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal of
Volume 201
Submit your manuscripts athttpswwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 201
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
4 Advances in Multimedia
Input audio X andwatermark w
Divide audio into framessized S samples
Calculate the statisticfeatures value E of the
frames
Set threshold T(T gt max(abs(E))) and
threshold G (usually G gt T)
wk = 1
No change
Combine the frames to obtainthe watermarked audio
End
Ek isin [0 T]
Yes
Yes
Some samples in the frameare added by value 훽 and
E㰀k = Ek + T + G
No
No
Some samples in the frameare subtracted by value 훽 and
E㰀k = Ek minus T minus G
Figure 4 Watermark embedding process
3 Proposed Algorithm
The embedding and extraction processes are presented indetail as follows
31 Watermark Embedding Figure 4 shows the proposedwatermark embedding process The watermark is embeddedwith the following five steps
Step 1 Divide the original audio 119883 into nonoverlappingframes sized 119878 samples
Step 2 Calculate the statistic features of the frames (119864 values)by referring to (1)ndash(3)
Step 3 Set the threshold values 119879 and 119866 (119879 gt 119864max andusually 119866 gt 119879)Step 4 If the watermark bit is ldquo0rdquo nothing is changed Ifthe bit is ldquo1rdquo shift the statistic feature value with a shiftingquantity 119879 +119866 to embed the watermark bit by modifying thesamples in the frame with value 120573 referring to (4)Step 5 Combine the frames to get the watermarked audio
32 Watermark Extraction If the watermarked audio goesthrough some attacks (such as MP3 compression additivenoise resampling or requantization) the watermark can
still be detected To improve the accuracy of the watermarkextraction three extraction methods and a majority votingsystem are adopted to identify the extracted watermark bycomputing the distorted statistical feature values 1198641015840(i) Extraction 1 Redefine the bit-0 region as [minus119879 minus 1198662 119879 +1198662] and the watermark extraction as
11990810158401 (119894) = 0 if1198641015840 isin [minus119879 minus 1198662 119879 + 1198662]1 otherwise (8)
(ii) Extraction 2 Redefine the bit-0 region as [minus119879 minus 1198663 119879 +1198663] and the watermark extraction as
11990810158402 (119894) = 0 if1198641015840 isin [minus119879 minus 1198663 119879 + 1198663]1 otherwise (9)
(iii) Extraction 3 119896-means clustering algorithm is introducedto extract bits Figure 5 shows the distribution of the 119864 valuesafter MP3 compression and the watermark can be extractedby
11990810158403 (119894) = 0 if1198641015840 isin class 21 if1198641015840 isin class 1 and class 3 (10)
Advances in Multimedia 5
Class 1 Class 3
Class 2
3000
2000
10000
4000
5000
minus20
00
minus30
00
minus40
00
minus10
00
minus50
00
02468
101214161820
Figure 5 The distribution of the statistic feature values of track 1after MP3 compression at 64 kbps
The majority voting system works as
1199081015840 (119894) =
11990810158401 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158401 (119894) = 11990810158403 (119894)11990810158401 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158401 (119894) = 11990810158403 (119894)11990810158402 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158402 (119894) = 11990810158403 (119894)11990810158403 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158401 (119894) = 11990810158403 (119894)
(11)
Eventually three extraction methods and a majorityvoting system are adopted to extract watermark Figure 6shows the proposed watermark extracting process If thewatermarked audio remains intact the watermark can beextracted correctly and the original audio can be recoveredas the following steps
Step 1 Divide the watermarked audio 119883119908 into nonoverlap-ping frames sized 119878 samples
Step 2 Calculate the statistic feature of the frames (1198641015840 values)by referring to (1)ndash(3)
Step 3 Extract the watermark with three extraction methodsand identify the watermark with the majority voting systemby referring to (8)ndash(11)
Step 4 The original audio can be recovered by modifying thesamples in the frame with value 120573 referring to (7)Step 5 Combine the frames to get the original audio
If the watermarked audio goes through some attacks theoriginal audio cannot be recovered exactly so we focus onthe watermark extraction and the watermark is extracted asfollows
Step 1 Divide the watermarked audio 119883119908 into nonoverlap-ping frames sized 119878 samples
Step 2 Calculate the statistic feature values of the frames 1198641015840by referring to (1)ndash(3)
Step 3 Extract the watermark with three extraction methodsand identify the watermark with the majority voting systemby referring to (8)ndash(11)
4 Experimental Results
In this section 7WAVaudio file of the sample rate of 441 KHzand 16 bits per sample (tracks 1 2 3 4 5 6 and 7 [18]) are usedas example clips to evaluate the performance of the proposedalgorithm The payload of our method only depends on thelength of a frame 119878 for a time-discrete digital audio signal119883in length119872 the pure payload can be calculated by
119862 = lfloor119872119878 rfloor (12)
In the experiment the watermark is a pseudo-randomsequence in length of 1000 bits The imperceptibility is firstanalyzed by the SNR standard at different threshold valuesand different sample numbers per frame Then robustnesstesting against MP3 compression additive noise (AWGN)resampling (441-16-441 kHz) and requantization (16-8-16bits) are reported by using the software CoolEditPro v21
41 Imperceptibility Test The imperceptibility is measuredby the embedding distortion In the proposed scheme thedistortion is caused by the shifting quantity on the samplesdepending on thresholds 119879 119866 and the length of a frame 119878Since 119879 is set at first we only investigate the influence of 119866and 119878 on SNR
Figure 7 plots the relationship between SNR and thethreshold 119866 for different clips at the same threshold 119879 and119878 From this figure we can conclude that with the increase of119866 the SNR value drops The reason is that the larger 119866 isthe larger shifting quantity is used so the larger embeddingdistortion is caused As a result SNR value drops
Figure 8 plots the relationship between SNR and thelength of a frame 119878 for different clips at the same thresholds119879 and 119866 We can see from this figure that the larger 119878is the higher SNR value is achieved The reason is thatwith the increase of 119878 the shifting quantity for every singlesample drops so the SNR values rise due to the fact that theembedding distortion is reduced Consequently the framelength 119878 will influence Maximum Embedding Capacity andSNR value directly the Maximum Embedding Capacity ishigher when 119878 is smaller according to (12) and the SNR valueis higher when 119878 is larger according to Figure 8 To considerthe balance between Maximum Embedding Capacity andSNR value we have found that the value of 119878within the rangeof 300 to 600 can usually achieve a satisfactory result after aset of experiments
42 Robustness Testing To test the robustness of the pro-posed scheme a set of experiments has been taken ontracks 1ndash7 Table 1 shows the results in which RP meansresampling (441-16-441 kHz) operations while RQ meansrequantization (16-8-16 bits) operations We can observefrom this table that all the example clips can achieve therobustness against MP3 compression at 64Kbps For track1 the watermark bits can be correctly extracted under the
6 Advances in Multimedia
Input audio Xw
Divide audio into framessized S samples
Calculate the statisticfeatures value E㰀 of the
frames
Three extraction methodsand a majority voting system
Watermark w
End
Yes
Yes
No
No
E㰀k isin [T + G 2T + G]
No change
E㰀k isin [minus2T minus G minusT minus G]
Combine the frames toobtain the original audio
Some samples in the frameare subtracted by value 훽 and
Ek = E㰀k minus T minus G
Some samples in the frameare added by value 훽 and
Ek = E㰀k + T + G
Figure 6 Watermark extracting process
3500 4000 4500 5000 5500 6000 6500 7000 75003000Threshold G
40
45
50
55
60
SNR
(dB)
Track 1 T = 500 S = 300Track 2 T = 5300 S = 300Track 3 T = 300 S = 300Track 4 T = 9900 S = 300
Track 5 T = 200 S = 300Track 6 T = 200 S = 300Track 7 T = 1700 S = 300
Figure 7 Relationship between SNR and threshold 119866
MP3 compression of 48KbpsThe robustness against additivenoise is also satisfactory Even with the noise intensity at25 dB the BER (bit error rate) values are less than 10except for track 1 Besides the watermark robustness againstresampling and requantization operations is perfect and thehidden bits can be recovered without errors
350 400 450 500 550 600300Length S
1015202530354045505560
SNR
(dB)
Track 1 T = 550G = 3000Track 2 T = 6200G = 7000Track 3 T = 1700G = 4000Track 4 T = 16000G = 17000
Track 5 T = 300G = 3000Track 6 T = 200G = 3000Track 7 T = 2700G = 3000
Figure 8 Relationship between SNR and the frame length 119878
As shown in Figure 3 the robustness of the proposedmethod is originated from the robust region The robustregion depends on threshold 119866 The larger 119866 is the largerthe robust region is and the stronger robustness is Figure 9
Advances in Multimedia 7
Table 1 Performance of the proposed method
Audio 119878 119879 119866 SNR MP3 (Kbps) AWGN (35 dB) AWGN (30 dB) AWGN (25 dB) RP RQTrack 1 420 500 3000 5474 48 21000 771000 2511000 01000 01000Track 2 600 6200 7000 4798 64 01000 01000 01000 01000 01000Track 3 510 600 3000 5409 48 01000 01000 441000 01000 01000Track 4 300 9900 10000 3899 64 01000 01000 01000 01000 01000Track 5 300 200 3000 5618 48 01000 251000 771000 01000 01000Track 6 600 200 3000 495 48 01000 01000 461000 01000 01000Track 7 300 1600 4000 4378 48 01000 01000 01000 01000 01000
Table 2 Performance of the proposed method with different length of a frame 119878 on track 1
Track S T G MP3 AWGN AWGN AWGN(Kbps) (35 dB) (30 dB) (25 dB)
Track 1 300 500 3000 80 91000 901000 2991000Track 1 420 500 3000 56 121000 1711000 3201000Track 1 510 550 3000 80 281000 1711000 1581000Track 1 600 550 3000 48 281000 2211000 3901000Track 6 300 200 3000 48 01000 01000 31000Track 6 420 200 3000 56 281000 01000 231000Track 7 300 200 4000 48 01000 01000 01000Track 7 420 200 4000 56 01000 01000 01000
Track 1 G = 5000Track 1 G = 6000Track 2 G = 6000Track 2 G = 7000
Track 3 G = 3000Track 3 G = 4000Track 4 G = 9000Track 4 G = 10000
50 55 60 65 70 75 80 85 90 95 10045MP3 compression bit rate
0
001
002
003
004
005
Bit e
rror
rate
(BER
)
Figure 9 Relationship between MP3 bit rate and threshold 119866
supports the conclusion Figure 9 shows the bit error rate(BER) at different threshold 119866 for the same audio withsame threshold 119879 The lower BER means that the strongerrobustness is achieved We can find that as threshold 119866increases the BER drops and the robustness rises
Take track 1 as example clip Figure 10 shows the bit errorrate of the extracted watermark with different thresholds 119866against additive noises with the same 119879 and 119878 (119879 = 500 119878 =300) We can see that the larger 119866 is the smaller bit errorrate is and the better robustness is As threshold 119866 increasesthe robustness becomes stronger In the application we canadjust the parameter 119866 to achieve ideal robustness On the
25 30 35 4020AWGN (dB)
0005
01015
02025
03035
04045
05
Bit e
rror
rate
(BER
)
Track 1 G = 3000Track 1 G = 4000Track 1 G = 5000
Track 1 G = 6000Track 1 G = 7500
Figure 10 The BER values at different AWGN with differentthreshold 119866
other hand with the increase of 119866 the SNR value drops Toconsider the balance between SNR value and robustness wehave found that the value of119866within 3000 to 5000 can usuallyachieve a satisfactory result after a set of experiments
To evaluate the effect of the frame length 119878 on therobustness performance a set of experiments has been takenon track 1 track 6 and track 7 Table 2 lists the results Wecan observe that for the same audio with the same 119879 and 119866such as 119879 = 500 and 119866 = 3000 for track 1 as 119878 increasesthe robustness against MP3 compression will strengthen butfor track 6 and track 7 as 119878 increases the robustness against
8 Advances in Multimedia
Table3Com
paris
onwith
metho
din
[17]
Track
Metho
din
[17]
Prop
osed
metho
dPayload
ODG
MP3
AWGN
119878119879
119866Payload
ODG
MP3
MP3
AWGN
(bits)
(128
Kbps)
(36d
B)(bits)
(80K
bps)
(64K
bps)
(36d
B)Track32
216
minus245
00
150
1500
3000
1000
minus137
002
0Track35
216
minus131
012
150
3500
5000
1000
minus158
008
0Track65
216
minus121
06
150
302000
1000
minus023
001
12
Track66
216
minus024
06
9660
0060
001000
minus105
03
12
0Track69
216
minus031
14
300
400
2000
1000
minus023
00
0
Advances in Multimedia 9
MP3 compression drops so the effect of the frame length 119878on the robustness against MP3 compression is unstable Theinfluence on AWGN is little
For fair comparison with the method in [17] we use thesame host signals (tracks 32 35 65 66 and 69) downloadedfrom sound quality assessment material (SQAM) collection[19] Table 3 shows the robustness testing results against MP3compression and additive noise (AWGN) operations We canobserve that the method in [17] can carry 216 bits and resisttheMP3 compression at 128Kbpswhile the proposedmethodcan resist the MP3 compression at 64Kbps with 1000 bitsembedded In addition in our method the BER under theAWGN of 35 dB is less than that of the method in [17]In other words the proposed method can provide largerembedding capacity and obtain stronger robustness againstMP3 compression and AWGN attacks The imperceptibilityis evaluated by using the ODG standardThe closer the ODGvalue to 0 the better the imperceptibility For the table it isnoted that the imperceptibility of the proposed method isbetter except for the clips track 35 and track 66 The reason isthat119864max values of track 35 and track 66 are bigger As a resultthresholds 119879 and 119866 are also larger and more embeddingdistortion will be caused
5 Conclusions
In this paper we proposed a robust and reversible audiowatermarking method by shifting the histogram of thestatistical feature values in time domainThe statistical featureis calculated as the sum of the prediction errors in a frameSince the audio clip has a larger number of samples andeach frame can hold enough elements the statistical featureis robust to those common signal processing operationsConsidering that the distribution of the statistical featurevalues may be distorted to some extent three extractionmethods and the majority voting system are designed forthe watermark detection Experimental results have shownthat thousands of bits can be reversibly embedded and thewatermark bits can resist MP3 compression of 64 kbps andadditive noise of 35 dB Comparingwith the existing excellentmethod the proposed method can embed more watermarkbits and achieve stronger robustness
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This work was supported by NSFC (no 61272414) and OpenResearch Fund from the State Key Laboratory of InformationSecurity (no 2016-MS-07)
References
[1] Y Xiang I Natgunanathan S GuoW Zhou and S NahavandildquoPatchwork-based audio watermarking method robust to de-synchronization attacksrdquo IEEE Transactions on Audio Speechand Language Processing vol 22 no 9 pp 1413ndash1423 2014
[2] J Fridrich M Goljan and R Du ldquoLossless data embeddingmdashnew paradigm in digital watermarkingrdquo EURASIP Journal onAdvances in Signal Processing vol 2002 no 2 pp 185ndash196 2002
[3] Y Q Shi Z Ni D Zou C Liang and G Xuan ldquoLosslessdata hiding fundamentals algorithms and applicationsrdquo inProceeding of the IEEE International Symposium on Circuits andSystems vol 2 pp 313ndash336 2004
[4] S Lee CD Yoo andTKalker ldquoReversible imagewatermarkingbased on integer-to-integer wavelet transformrdquo IEEE Transac-tions on Information Forensics and Security vol 2 no 3 pp 321ndash330 2010
[5] X Li B Yang and T Zeng ldquoEfficient reversible watermarkingbased on adaptive prediction-error expansion and pixel selec-tionrdquo IEEE Transactions on Image Processing vol 20 no 12 pp3524ndash3533 2000
[6] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hidingrdquo in Proceeding of the IEEE Interna-tional Conference onMultimedia and Expo (ICME rsquo2004) vol 3pp 2199ndash2202 Taipei Taiwan June 2004
[7] C De Vleeschouwer J Delaigle and B Macq ldquoCircularinterpretation of histogram for reversible watermarkingrdquo inProceeding of the IEEE 4th Workshop on Multimedia SignalProcessing pp 345ndash350 Cannes France 2001
[8] C De Vleeschouwer J F Delaigle and B Macq ldquoCircularinterpretation of bijective transformations in lossless water-marking for media asset managementrdquo IEEE Transactions onMultimedia vol 5 no 1 pp 97ndash105 2003
[9] D Zou YQ Shi ZNi andW Su ldquoA semi-fragile lossless digitalwatermarking scheme based on integer wavelet transformrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 10 pp 1294ndash1300 2006
[10] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hiding designed for semi-fragile imageauthenticationrdquo IEEE Transactions on Circuits and Systems forVideo Technology vol 18 no 4 pp 890ndash896 2008
[11] L An X Gao C Deng and F Ji ldquoRobust lossless data hidingAnalysis and evaluationrdquo in Procceding of the InternationalConference on High Performance Computing and Simulation(HPCS rsquo10) pp 512ndash516 July 2010
[12] X-T Zeng L-D Ping and X-Z Pan ldquoA lossless robust datahiding schemerdquo Pattern Recognition vol 43 no 4 pp 1656ndash1667 2010
[13] L An X Gao X Li D Tao C Deng and J Li ldquoRobustreversible watermarking via clustering and enhanced pixel-wisemaskingrdquo IEEE Transactions on Image Processing vol 21 no 8pp 3598ndash3611 2012
[14] R Thabit and B E Khoo ldquoCapacity improved robust losslessimage watermarkingrdquo IET Image Processing vol 8 no 11 pp662ndash670 2014
[15] S Xiang and Y Wang ldquoDistortion-free robust reversible water-marking by modifying and recording iwt means of imageblocksrdquo in Proceeding of the 14th International Workshop(IWDW rsquo15) Tokyo Japan October 2015
[16] D Coltuc and J Chassery ldquoDistortion-free robust watermark-ing a case studyrdquo in Security Steganography andWatermarkingof Multimedia Contents vol 6505 of Proceedings of SPIE pp588ndash595 San Jose Calif USA 2007
[17] A Nishimura ldquoReversible and robust audio watermarkingbased on spread spectrum and amplitude expansionrdquo in Inter-national Workshop on Digital Watermarking (IWDW rsquo14) vol9023 of Lecture Notes in Computer Science pp 215ndash229
10 Advances in Multimedia
[18] Massachusetts Institute of Technology (MIT) Audio Databasehttpsoundmediamitedumediaphp
[19] EBUCommittee Sound quality assessmentmaterial recordingsfor subjective tests httpstechebuchpublicationssqamcd
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal of
Volume 201
Submit your manuscripts athttpswwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 201
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
Advances in Multimedia 5
Class 1 Class 3
Class 2
3000
2000
10000
4000
5000
minus20
00
minus30
00
minus40
00
minus10
00
minus50
00
02468
101214161820
Figure 5 The distribution of the statistic feature values of track 1after MP3 compression at 64 kbps
The majority voting system works as
1199081015840 (119894) =
11990810158401 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158401 (119894) = 11990810158403 (119894)11990810158401 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158401 (119894) = 11990810158403 (119894)11990810158402 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158402 (119894) = 11990810158403 (119894)11990810158403 (119894) if11990810158401 (119894) = 11990810158402 (119894) 11990810158401 (119894) = 11990810158403 (119894)
(11)
Eventually three extraction methods and a majorityvoting system are adopted to extract watermark Figure 6shows the proposed watermark extracting process If thewatermarked audio remains intact the watermark can beextracted correctly and the original audio can be recoveredas the following steps
Step 1 Divide the watermarked audio 119883119908 into nonoverlap-ping frames sized 119878 samples
Step 2 Calculate the statistic feature of the frames (1198641015840 values)by referring to (1)ndash(3)
Step 3 Extract the watermark with three extraction methodsand identify the watermark with the majority voting systemby referring to (8)ndash(11)
Step 4 The original audio can be recovered by modifying thesamples in the frame with value 120573 referring to (7)Step 5 Combine the frames to get the original audio
If the watermarked audio goes through some attacks theoriginal audio cannot be recovered exactly so we focus onthe watermark extraction and the watermark is extracted asfollows
Step 1 Divide the watermarked audio 119883119908 into nonoverlap-ping frames sized 119878 samples
Step 2 Calculate the statistic feature values of the frames 1198641015840by referring to (1)ndash(3)
Step 3 Extract the watermark with three extraction methodsand identify the watermark with the majority voting systemby referring to (8)ndash(11)
4 Experimental Results
In this section 7WAVaudio file of the sample rate of 441 KHzand 16 bits per sample (tracks 1 2 3 4 5 6 and 7 [18]) are usedas example clips to evaluate the performance of the proposedalgorithm The payload of our method only depends on thelength of a frame 119878 for a time-discrete digital audio signal119883in length119872 the pure payload can be calculated by
119862 = lfloor119872119878 rfloor (12)
In the experiment the watermark is a pseudo-randomsequence in length of 1000 bits The imperceptibility is firstanalyzed by the SNR standard at different threshold valuesand different sample numbers per frame Then robustnesstesting against MP3 compression additive noise (AWGN)resampling (441-16-441 kHz) and requantization (16-8-16bits) are reported by using the software CoolEditPro v21
41 Imperceptibility Test The imperceptibility is measuredby the embedding distortion In the proposed scheme thedistortion is caused by the shifting quantity on the samplesdepending on thresholds 119879 119866 and the length of a frame 119878Since 119879 is set at first we only investigate the influence of 119866and 119878 on SNR
Figure 7 plots the relationship between SNR and thethreshold 119866 for different clips at the same threshold 119879 and119878 From this figure we can conclude that with the increase of119866 the SNR value drops The reason is that the larger 119866 isthe larger shifting quantity is used so the larger embeddingdistortion is caused As a result SNR value drops
Figure 8 plots the relationship between SNR and thelength of a frame 119878 for different clips at the same thresholds119879 and 119866 We can see from this figure that the larger 119878is the higher SNR value is achieved The reason is thatwith the increase of 119878 the shifting quantity for every singlesample drops so the SNR values rise due to the fact that theembedding distortion is reduced Consequently the framelength 119878 will influence Maximum Embedding Capacity andSNR value directly the Maximum Embedding Capacity ishigher when 119878 is smaller according to (12) and the SNR valueis higher when 119878 is larger according to Figure 8 To considerthe balance between Maximum Embedding Capacity andSNR value we have found that the value of 119878within the rangeof 300 to 600 can usually achieve a satisfactory result after aset of experiments
42 Robustness Testing To test the robustness of the pro-posed scheme a set of experiments has been taken ontracks 1ndash7 Table 1 shows the results in which RP meansresampling (441-16-441 kHz) operations while RQ meansrequantization (16-8-16 bits) operations We can observefrom this table that all the example clips can achieve therobustness against MP3 compression at 64Kbps For track1 the watermark bits can be correctly extracted under the
6 Advances in Multimedia
Input audio Xw
Divide audio into framessized S samples
Calculate the statisticfeatures value E㰀 of the
frames
Three extraction methodsand a majority voting system
Watermark w
End
Yes
Yes
No
No
E㰀k isin [T + G 2T + G]
No change
E㰀k isin [minus2T minus G minusT minus G]
Combine the frames toobtain the original audio
Some samples in the frameare subtracted by value 훽 and
Ek = E㰀k minus T minus G
Some samples in the frameare added by value 훽 and
Ek = E㰀k + T + G
Figure 6 Watermark extracting process
3500 4000 4500 5000 5500 6000 6500 7000 75003000Threshold G
40
45
50
55
60
SNR
(dB)
Track 1 T = 500 S = 300Track 2 T = 5300 S = 300Track 3 T = 300 S = 300Track 4 T = 9900 S = 300
Track 5 T = 200 S = 300Track 6 T = 200 S = 300Track 7 T = 1700 S = 300
Figure 7 Relationship between SNR and threshold 119866
MP3 compression of 48KbpsThe robustness against additivenoise is also satisfactory Even with the noise intensity at25 dB the BER (bit error rate) values are less than 10except for track 1 Besides the watermark robustness againstresampling and requantization operations is perfect and thehidden bits can be recovered without errors
350 400 450 500 550 600300Length S
1015202530354045505560
SNR
(dB)
Track 1 T = 550G = 3000Track 2 T = 6200G = 7000Track 3 T = 1700G = 4000Track 4 T = 16000G = 17000
Track 5 T = 300G = 3000Track 6 T = 200G = 3000Track 7 T = 2700G = 3000
Figure 8 Relationship between SNR and the frame length 119878
As shown in Figure 3 the robustness of the proposedmethod is originated from the robust region The robustregion depends on threshold 119866 The larger 119866 is the largerthe robust region is and the stronger robustness is Figure 9
Advances in Multimedia 7
Table 1 Performance of the proposed method
Audio 119878 119879 119866 SNR MP3 (Kbps) AWGN (35 dB) AWGN (30 dB) AWGN (25 dB) RP RQTrack 1 420 500 3000 5474 48 21000 771000 2511000 01000 01000Track 2 600 6200 7000 4798 64 01000 01000 01000 01000 01000Track 3 510 600 3000 5409 48 01000 01000 441000 01000 01000Track 4 300 9900 10000 3899 64 01000 01000 01000 01000 01000Track 5 300 200 3000 5618 48 01000 251000 771000 01000 01000Track 6 600 200 3000 495 48 01000 01000 461000 01000 01000Track 7 300 1600 4000 4378 48 01000 01000 01000 01000 01000
Table 2 Performance of the proposed method with different length of a frame 119878 on track 1
Track S T G MP3 AWGN AWGN AWGN(Kbps) (35 dB) (30 dB) (25 dB)
Track 1 300 500 3000 80 91000 901000 2991000Track 1 420 500 3000 56 121000 1711000 3201000Track 1 510 550 3000 80 281000 1711000 1581000Track 1 600 550 3000 48 281000 2211000 3901000Track 6 300 200 3000 48 01000 01000 31000Track 6 420 200 3000 56 281000 01000 231000Track 7 300 200 4000 48 01000 01000 01000Track 7 420 200 4000 56 01000 01000 01000
Track 1 G = 5000Track 1 G = 6000Track 2 G = 6000Track 2 G = 7000
Track 3 G = 3000Track 3 G = 4000Track 4 G = 9000Track 4 G = 10000
50 55 60 65 70 75 80 85 90 95 10045MP3 compression bit rate
0
001
002
003
004
005
Bit e
rror
rate
(BER
)
Figure 9 Relationship between MP3 bit rate and threshold 119866
supports the conclusion Figure 9 shows the bit error rate(BER) at different threshold 119866 for the same audio withsame threshold 119879 The lower BER means that the strongerrobustness is achieved We can find that as threshold 119866increases the BER drops and the robustness rises
Take track 1 as example clip Figure 10 shows the bit errorrate of the extracted watermark with different thresholds 119866against additive noises with the same 119879 and 119878 (119879 = 500 119878 =300) We can see that the larger 119866 is the smaller bit errorrate is and the better robustness is As threshold 119866 increasesthe robustness becomes stronger In the application we canadjust the parameter 119866 to achieve ideal robustness On the
25 30 35 4020AWGN (dB)
0005
01015
02025
03035
04045
05
Bit e
rror
rate
(BER
)
Track 1 G = 3000Track 1 G = 4000Track 1 G = 5000
Track 1 G = 6000Track 1 G = 7500
Figure 10 The BER values at different AWGN with differentthreshold 119866
other hand with the increase of 119866 the SNR value drops Toconsider the balance between SNR value and robustness wehave found that the value of119866within 3000 to 5000 can usuallyachieve a satisfactory result after a set of experiments
To evaluate the effect of the frame length 119878 on therobustness performance a set of experiments has been takenon track 1 track 6 and track 7 Table 2 lists the results Wecan observe that for the same audio with the same 119879 and 119866such as 119879 = 500 and 119866 = 3000 for track 1 as 119878 increasesthe robustness against MP3 compression will strengthen butfor track 6 and track 7 as 119878 increases the robustness against
8 Advances in Multimedia
Table3Com
paris
onwith
metho
din
[17]
Track
Metho
din
[17]
Prop
osed
metho
dPayload
ODG
MP3
AWGN
119878119879
119866Payload
ODG
MP3
MP3
AWGN
(bits)
(128
Kbps)
(36d
B)(bits)
(80K
bps)
(64K
bps)
(36d
B)Track32
216
minus245
00
150
1500
3000
1000
minus137
002
0Track35
216
minus131
012
150
3500
5000
1000
minus158
008
0Track65
216
minus121
06
150
302000
1000
minus023
001
12
Track66
216
minus024
06
9660
0060
001000
minus105
03
12
0Track69
216
minus031
14
300
400
2000
1000
minus023
00
0
Advances in Multimedia 9
MP3 compression drops so the effect of the frame length 119878on the robustness against MP3 compression is unstable Theinfluence on AWGN is little
For fair comparison with the method in [17] we use thesame host signals (tracks 32 35 65 66 and 69) downloadedfrom sound quality assessment material (SQAM) collection[19] Table 3 shows the robustness testing results against MP3compression and additive noise (AWGN) operations We canobserve that the method in [17] can carry 216 bits and resisttheMP3 compression at 128Kbpswhile the proposedmethodcan resist the MP3 compression at 64Kbps with 1000 bitsembedded In addition in our method the BER under theAWGN of 35 dB is less than that of the method in [17]In other words the proposed method can provide largerembedding capacity and obtain stronger robustness againstMP3 compression and AWGN attacks The imperceptibilityis evaluated by using the ODG standardThe closer the ODGvalue to 0 the better the imperceptibility For the table it isnoted that the imperceptibility of the proposed method isbetter except for the clips track 35 and track 66 The reason isthat119864max values of track 35 and track 66 are bigger As a resultthresholds 119879 and 119866 are also larger and more embeddingdistortion will be caused
5 Conclusions
In this paper we proposed a robust and reversible audiowatermarking method by shifting the histogram of thestatistical feature values in time domainThe statistical featureis calculated as the sum of the prediction errors in a frameSince the audio clip has a larger number of samples andeach frame can hold enough elements the statistical featureis robust to those common signal processing operationsConsidering that the distribution of the statistical featurevalues may be distorted to some extent three extractionmethods and the majority voting system are designed forthe watermark detection Experimental results have shownthat thousands of bits can be reversibly embedded and thewatermark bits can resist MP3 compression of 64 kbps andadditive noise of 35 dB Comparingwith the existing excellentmethod the proposed method can embed more watermarkbits and achieve stronger robustness
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This work was supported by NSFC (no 61272414) and OpenResearch Fund from the State Key Laboratory of InformationSecurity (no 2016-MS-07)
References
[1] Y Xiang I Natgunanathan S GuoW Zhou and S NahavandildquoPatchwork-based audio watermarking method robust to de-synchronization attacksrdquo IEEE Transactions on Audio Speechand Language Processing vol 22 no 9 pp 1413ndash1423 2014
[2] J Fridrich M Goljan and R Du ldquoLossless data embeddingmdashnew paradigm in digital watermarkingrdquo EURASIP Journal onAdvances in Signal Processing vol 2002 no 2 pp 185ndash196 2002
[3] Y Q Shi Z Ni D Zou C Liang and G Xuan ldquoLosslessdata hiding fundamentals algorithms and applicationsrdquo inProceeding of the IEEE International Symposium on Circuits andSystems vol 2 pp 313ndash336 2004
[4] S Lee CD Yoo andTKalker ldquoReversible imagewatermarkingbased on integer-to-integer wavelet transformrdquo IEEE Transac-tions on Information Forensics and Security vol 2 no 3 pp 321ndash330 2010
[5] X Li B Yang and T Zeng ldquoEfficient reversible watermarkingbased on adaptive prediction-error expansion and pixel selec-tionrdquo IEEE Transactions on Image Processing vol 20 no 12 pp3524ndash3533 2000
[6] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hidingrdquo in Proceeding of the IEEE Interna-tional Conference onMultimedia and Expo (ICME rsquo2004) vol 3pp 2199ndash2202 Taipei Taiwan June 2004
[7] C De Vleeschouwer J Delaigle and B Macq ldquoCircularinterpretation of histogram for reversible watermarkingrdquo inProceeding of the IEEE 4th Workshop on Multimedia SignalProcessing pp 345ndash350 Cannes France 2001
[8] C De Vleeschouwer J F Delaigle and B Macq ldquoCircularinterpretation of bijective transformations in lossless water-marking for media asset managementrdquo IEEE Transactions onMultimedia vol 5 no 1 pp 97ndash105 2003
[9] D Zou YQ Shi ZNi andW Su ldquoA semi-fragile lossless digitalwatermarking scheme based on integer wavelet transformrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 10 pp 1294ndash1300 2006
[10] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hiding designed for semi-fragile imageauthenticationrdquo IEEE Transactions on Circuits and Systems forVideo Technology vol 18 no 4 pp 890ndash896 2008
[11] L An X Gao C Deng and F Ji ldquoRobust lossless data hidingAnalysis and evaluationrdquo in Procceding of the InternationalConference on High Performance Computing and Simulation(HPCS rsquo10) pp 512ndash516 July 2010
[12] X-T Zeng L-D Ping and X-Z Pan ldquoA lossless robust datahiding schemerdquo Pattern Recognition vol 43 no 4 pp 1656ndash1667 2010
[13] L An X Gao X Li D Tao C Deng and J Li ldquoRobustreversible watermarking via clustering and enhanced pixel-wisemaskingrdquo IEEE Transactions on Image Processing vol 21 no 8pp 3598ndash3611 2012
[14] R Thabit and B E Khoo ldquoCapacity improved robust losslessimage watermarkingrdquo IET Image Processing vol 8 no 11 pp662ndash670 2014
[15] S Xiang and Y Wang ldquoDistortion-free robust reversible water-marking by modifying and recording iwt means of imageblocksrdquo in Proceeding of the 14th International Workshop(IWDW rsquo15) Tokyo Japan October 2015
[16] D Coltuc and J Chassery ldquoDistortion-free robust watermark-ing a case studyrdquo in Security Steganography andWatermarkingof Multimedia Contents vol 6505 of Proceedings of SPIE pp588ndash595 San Jose Calif USA 2007
[17] A Nishimura ldquoReversible and robust audio watermarkingbased on spread spectrum and amplitude expansionrdquo in Inter-national Workshop on Digital Watermarking (IWDW rsquo14) vol9023 of Lecture Notes in Computer Science pp 215ndash229
10 Advances in Multimedia
[18] Massachusetts Institute of Technology (MIT) Audio Databasehttpsoundmediamitedumediaphp
[19] EBUCommittee Sound quality assessmentmaterial recordingsfor subjective tests httpstechebuchpublicationssqamcd
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal of
Volume 201
Submit your manuscripts athttpswwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 201
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
6 Advances in Multimedia
Input audio Xw
Divide audio into framessized S samples
Calculate the statisticfeatures value E㰀 of the
frames
Three extraction methodsand a majority voting system
Watermark w
End
Yes
Yes
No
No
E㰀k isin [T + G 2T + G]
No change
E㰀k isin [minus2T minus G minusT minus G]
Combine the frames toobtain the original audio
Some samples in the frameare subtracted by value 훽 and
Ek = E㰀k minus T minus G
Some samples in the frameare added by value 훽 and
Ek = E㰀k + T + G
Figure 6 Watermark extracting process
3500 4000 4500 5000 5500 6000 6500 7000 75003000Threshold G
40
45
50
55
60
SNR
(dB)
Track 1 T = 500 S = 300Track 2 T = 5300 S = 300Track 3 T = 300 S = 300Track 4 T = 9900 S = 300
Track 5 T = 200 S = 300Track 6 T = 200 S = 300Track 7 T = 1700 S = 300
Figure 7 Relationship between SNR and threshold 119866
MP3 compression of 48KbpsThe robustness against additivenoise is also satisfactory Even with the noise intensity at25 dB the BER (bit error rate) values are less than 10except for track 1 Besides the watermark robustness againstresampling and requantization operations is perfect and thehidden bits can be recovered without errors
350 400 450 500 550 600300Length S
1015202530354045505560
SNR
(dB)
Track 1 T = 550G = 3000Track 2 T = 6200G = 7000Track 3 T = 1700G = 4000Track 4 T = 16000G = 17000
Track 5 T = 300G = 3000Track 6 T = 200G = 3000Track 7 T = 2700G = 3000
Figure 8 Relationship between SNR and the frame length 119878
As shown in Figure 3 the robustness of the proposedmethod is originated from the robust region The robustregion depends on threshold 119866 The larger 119866 is the largerthe robust region is and the stronger robustness is Figure 9
Advances in Multimedia 7
Table 1 Performance of the proposed method
Audio 119878 119879 119866 SNR MP3 (Kbps) AWGN (35 dB) AWGN (30 dB) AWGN (25 dB) RP RQTrack 1 420 500 3000 5474 48 21000 771000 2511000 01000 01000Track 2 600 6200 7000 4798 64 01000 01000 01000 01000 01000Track 3 510 600 3000 5409 48 01000 01000 441000 01000 01000Track 4 300 9900 10000 3899 64 01000 01000 01000 01000 01000Track 5 300 200 3000 5618 48 01000 251000 771000 01000 01000Track 6 600 200 3000 495 48 01000 01000 461000 01000 01000Track 7 300 1600 4000 4378 48 01000 01000 01000 01000 01000
Table 2 Performance of the proposed method with different length of a frame 119878 on track 1
Track S T G MP3 AWGN AWGN AWGN(Kbps) (35 dB) (30 dB) (25 dB)
Track 1 300 500 3000 80 91000 901000 2991000Track 1 420 500 3000 56 121000 1711000 3201000Track 1 510 550 3000 80 281000 1711000 1581000Track 1 600 550 3000 48 281000 2211000 3901000Track 6 300 200 3000 48 01000 01000 31000Track 6 420 200 3000 56 281000 01000 231000Track 7 300 200 4000 48 01000 01000 01000Track 7 420 200 4000 56 01000 01000 01000
Track 1 G = 5000Track 1 G = 6000Track 2 G = 6000Track 2 G = 7000
Track 3 G = 3000Track 3 G = 4000Track 4 G = 9000Track 4 G = 10000
50 55 60 65 70 75 80 85 90 95 10045MP3 compression bit rate
0
001
002
003
004
005
Bit e
rror
rate
(BER
)
Figure 9 Relationship between MP3 bit rate and threshold 119866
supports the conclusion Figure 9 shows the bit error rate(BER) at different threshold 119866 for the same audio withsame threshold 119879 The lower BER means that the strongerrobustness is achieved We can find that as threshold 119866increases the BER drops and the robustness rises
Take track 1 as example clip Figure 10 shows the bit errorrate of the extracted watermark with different thresholds 119866against additive noises with the same 119879 and 119878 (119879 = 500 119878 =300) We can see that the larger 119866 is the smaller bit errorrate is and the better robustness is As threshold 119866 increasesthe robustness becomes stronger In the application we canadjust the parameter 119866 to achieve ideal robustness On the
25 30 35 4020AWGN (dB)
0005
01015
02025
03035
04045
05
Bit e
rror
rate
(BER
)
Track 1 G = 3000Track 1 G = 4000Track 1 G = 5000
Track 1 G = 6000Track 1 G = 7500
Figure 10 The BER values at different AWGN with differentthreshold 119866
other hand with the increase of 119866 the SNR value drops Toconsider the balance between SNR value and robustness wehave found that the value of119866within 3000 to 5000 can usuallyachieve a satisfactory result after a set of experiments
To evaluate the effect of the frame length 119878 on therobustness performance a set of experiments has been takenon track 1 track 6 and track 7 Table 2 lists the results Wecan observe that for the same audio with the same 119879 and 119866such as 119879 = 500 and 119866 = 3000 for track 1 as 119878 increasesthe robustness against MP3 compression will strengthen butfor track 6 and track 7 as 119878 increases the robustness against
8 Advances in Multimedia
Table3Com
paris
onwith
metho
din
[17]
Track
Metho
din
[17]
Prop
osed
metho
dPayload
ODG
MP3
AWGN
119878119879
119866Payload
ODG
MP3
MP3
AWGN
(bits)
(128
Kbps)
(36d
B)(bits)
(80K
bps)
(64K
bps)
(36d
B)Track32
216
minus245
00
150
1500
3000
1000
minus137
002
0Track35
216
minus131
012
150
3500
5000
1000
minus158
008
0Track65
216
minus121
06
150
302000
1000
minus023
001
12
Track66
216
minus024
06
9660
0060
001000
minus105
03
12
0Track69
216
minus031
14
300
400
2000
1000
minus023
00
0
Advances in Multimedia 9
MP3 compression drops so the effect of the frame length 119878on the robustness against MP3 compression is unstable Theinfluence on AWGN is little
For fair comparison with the method in [17] we use thesame host signals (tracks 32 35 65 66 and 69) downloadedfrom sound quality assessment material (SQAM) collection[19] Table 3 shows the robustness testing results against MP3compression and additive noise (AWGN) operations We canobserve that the method in [17] can carry 216 bits and resisttheMP3 compression at 128Kbpswhile the proposedmethodcan resist the MP3 compression at 64Kbps with 1000 bitsembedded In addition in our method the BER under theAWGN of 35 dB is less than that of the method in [17]In other words the proposed method can provide largerembedding capacity and obtain stronger robustness againstMP3 compression and AWGN attacks The imperceptibilityis evaluated by using the ODG standardThe closer the ODGvalue to 0 the better the imperceptibility For the table it isnoted that the imperceptibility of the proposed method isbetter except for the clips track 35 and track 66 The reason isthat119864max values of track 35 and track 66 are bigger As a resultthresholds 119879 and 119866 are also larger and more embeddingdistortion will be caused
5 Conclusions
In this paper we proposed a robust and reversible audiowatermarking method by shifting the histogram of thestatistical feature values in time domainThe statistical featureis calculated as the sum of the prediction errors in a frameSince the audio clip has a larger number of samples andeach frame can hold enough elements the statistical featureis robust to those common signal processing operationsConsidering that the distribution of the statistical featurevalues may be distorted to some extent three extractionmethods and the majority voting system are designed forthe watermark detection Experimental results have shownthat thousands of bits can be reversibly embedded and thewatermark bits can resist MP3 compression of 64 kbps andadditive noise of 35 dB Comparingwith the existing excellentmethod the proposed method can embed more watermarkbits and achieve stronger robustness
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This work was supported by NSFC (no 61272414) and OpenResearch Fund from the State Key Laboratory of InformationSecurity (no 2016-MS-07)
References
[1] Y Xiang I Natgunanathan S GuoW Zhou and S NahavandildquoPatchwork-based audio watermarking method robust to de-synchronization attacksrdquo IEEE Transactions on Audio Speechand Language Processing vol 22 no 9 pp 1413ndash1423 2014
[2] J Fridrich M Goljan and R Du ldquoLossless data embeddingmdashnew paradigm in digital watermarkingrdquo EURASIP Journal onAdvances in Signal Processing vol 2002 no 2 pp 185ndash196 2002
[3] Y Q Shi Z Ni D Zou C Liang and G Xuan ldquoLosslessdata hiding fundamentals algorithms and applicationsrdquo inProceeding of the IEEE International Symposium on Circuits andSystems vol 2 pp 313ndash336 2004
[4] S Lee CD Yoo andTKalker ldquoReversible imagewatermarkingbased on integer-to-integer wavelet transformrdquo IEEE Transac-tions on Information Forensics and Security vol 2 no 3 pp 321ndash330 2010
[5] X Li B Yang and T Zeng ldquoEfficient reversible watermarkingbased on adaptive prediction-error expansion and pixel selec-tionrdquo IEEE Transactions on Image Processing vol 20 no 12 pp3524ndash3533 2000
[6] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hidingrdquo in Proceeding of the IEEE Interna-tional Conference onMultimedia and Expo (ICME rsquo2004) vol 3pp 2199ndash2202 Taipei Taiwan June 2004
[7] C De Vleeschouwer J Delaigle and B Macq ldquoCircularinterpretation of histogram for reversible watermarkingrdquo inProceeding of the IEEE 4th Workshop on Multimedia SignalProcessing pp 345ndash350 Cannes France 2001
[8] C De Vleeschouwer J F Delaigle and B Macq ldquoCircularinterpretation of bijective transformations in lossless water-marking for media asset managementrdquo IEEE Transactions onMultimedia vol 5 no 1 pp 97ndash105 2003
[9] D Zou YQ Shi ZNi andW Su ldquoA semi-fragile lossless digitalwatermarking scheme based on integer wavelet transformrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 10 pp 1294ndash1300 2006
[10] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hiding designed for semi-fragile imageauthenticationrdquo IEEE Transactions on Circuits and Systems forVideo Technology vol 18 no 4 pp 890ndash896 2008
[11] L An X Gao C Deng and F Ji ldquoRobust lossless data hidingAnalysis and evaluationrdquo in Procceding of the InternationalConference on High Performance Computing and Simulation(HPCS rsquo10) pp 512ndash516 July 2010
[12] X-T Zeng L-D Ping and X-Z Pan ldquoA lossless robust datahiding schemerdquo Pattern Recognition vol 43 no 4 pp 1656ndash1667 2010
[13] L An X Gao X Li D Tao C Deng and J Li ldquoRobustreversible watermarking via clustering and enhanced pixel-wisemaskingrdquo IEEE Transactions on Image Processing vol 21 no 8pp 3598ndash3611 2012
[14] R Thabit and B E Khoo ldquoCapacity improved robust losslessimage watermarkingrdquo IET Image Processing vol 8 no 11 pp662ndash670 2014
[15] S Xiang and Y Wang ldquoDistortion-free robust reversible water-marking by modifying and recording iwt means of imageblocksrdquo in Proceeding of the 14th International Workshop(IWDW rsquo15) Tokyo Japan October 2015
[16] D Coltuc and J Chassery ldquoDistortion-free robust watermark-ing a case studyrdquo in Security Steganography andWatermarkingof Multimedia Contents vol 6505 of Proceedings of SPIE pp588ndash595 San Jose Calif USA 2007
[17] A Nishimura ldquoReversible and robust audio watermarkingbased on spread spectrum and amplitude expansionrdquo in Inter-national Workshop on Digital Watermarking (IWDW rsquo14) vol9023 of Lecture Notes in Computer Science pp 215ndash229
10 Advances in Multimedia
[18] Massachusetts Institute of Technology (MIT) Audio Databasehttpsoundmediamitedumediaphp
[19] EBUCommittee Sound quality assessmentmaterial recordingsfor subjective tests httpstechebuchpublicationssqamcd
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal of
Volume 201
Submit your manuscripts athttpswwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 201
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
Advances in Multimedia 7
Table 1 Performance of the proposed method
Audio 119878 119879 119866 SNR MP3 (Kbps) AWGN (35 dB) AWGN (30 dB) AWGN (25 dB) RP RQTrack 1 420 500 3000 5474 48 21000 771000 2511000 01000 01000Track 2 600 6200 7000 4798 64 01000 01000 01000 01000 01000Track 3 510 600 3000 5409 48 01000 01000 441000 01000 01000Track 4 300 9900 10000 3899 64 01000 01000 01000 01000 01000Track 5 300 200 3000 5618 48 01000 251000 771000 01000 01000Track 6 600 200 3000 495 48 01000 01000 461000 01000 01000Track 7 300 1600 4000 4378 48 01000 01000 01000 01000 01000
Table 2 Performance of the proposed method with different length of a frame 119878 on track 1
Track S T G MP3 AWGN AWGN AWGN(Kbps) (35 dB) (30 dB) (25 dB)
Track 1 300 500 3000 80 91000 901000 2991000Track 1 420 500 3000 56 121000 1711000 3201000Track 1 510 550 3000 80 281000 1711000 1581000Track 1 600 550 3000 48 281000 2211000 3901000Track 6 300 200 3000 48 01000 01000 31000Track 6 420 200 3000 56 281000 01000 231000Track 7 300 200 4000 48 01000 01000 01000Track 7 420 200 4000 56 01000 01000 01000
Track 1 G = 5000Track 1 G = 6000Track 2 G = 6000Track 2 G = 7000
Track 3 G = 3000Track 3 G = 4000Track 4 G = 9000Track 4 G = 10000
50 55 60 65 70 75 80 85 90 95 10045MP3 compression bit rate
0
001
002
003
004
005
Bit e
rror
rate
(BER
)
Figure 9 Relationship between MP3 bit rate and threshold 119866
supports the conclusion Figure 9 shows the bit error rate(BER) at different threshold 119866 for the same audio withsame threshold 119879 The lower BER means that the strongerrobustness is achieved We can find that as threshold 119866increases the BER drops and the robustness rises
Take track 1 as example clip Figure 10 shows the bit errorrate of the extracted watermark with different thresholds 119866against additive noises with the same 119879 and 119878 (119879 = 500 119878 =300) We can see that the larger 119866 is the smaller bit errorrate is and the better robustness is As threshold 119866 increasesthe robustness becomes stronger In the application we canadjust the parameter 119866 to achieve ideal robustness On the
25 30 35 4020AWGN (dB)
0005
01015
02025
03035
04045
05
Bit e
rror
rate
(BER
)
Track 1 G = 3000Track 1 G = 4000Track 1 G = 5000
Track 1 G = 6000Track 1 G = 7500
Figure 10 The BER values at different AWGN with differentthreshold 119866
other hand with the increase of 119866 the SNR value drops Toconsider the balance between SNR value and robustness wehave found that the value of119866within 3000 to 5000 can usuallyachieve a satisfactory result after a set of experiments
To evaluate the effect of the frame length 119878 on therobustness performance a set of experiments has been takenon track 1 track 6 and track 7 Table 2 lists the results Wecan observe that for the same audio with the same 119879 and 119866such as 119879 = 500 and 119866 = 3000 for track 1 as 119878 increasesthe robustness against MP3 compression will strengthen butfor track 6 and track 7 as 119878 increases the robustness against
8 Advances in Multimedia
Table3Com
paris
onwith
metho
din
[17]
Track
Metho
din
[17]
Prop
osed
metho
dPayload
ODG
MP3
AWGN
119878119879
119866Payload
ODG
MP3
MP3
AWGN
(bits)
(128
Kbps)
(36d
B)(bits)
(80K
bps)
(64K
bps)
(36d
B)Track32
216
minus245
00
150
1500
3000
1000
minus137
002
0Track35
216
minus131
012
150
3500
5000
1000
minus158
008
0Track65
216
minus121
06
150
302000
1000
minus023
001
12
Track66
216
minus024
06
9660
0060
001000
minus105
03
12
0Track69
216
minus031
14
300
400
2000
1000
minus023
00
0
Advances in Multimedia 9
MP3 compression drops so the effect of the frame length 119878on the robustness against MP3 compression is unstable Theinfluence on AWGN is little
For fair comparison with the method in [17] we use thesame host signals (tracks 32 35 65 66 and 69) downloadedfrom sound quality assessment material (SQAM) collection[19] Table 3 shows the robustness testing results against MP3compression and additive noise (AWGN) operations We canobserve that the method in [17] can carry 216 bits and resisttheMP3 compression at 128Kbpswhile the proposedmethodcan resist the MP3 compression at 64Kbps with 1000 bitsembedded In addition in our method the BER under theAWGN of 35 dB is less than that of the method in [17]In other words the proposed method can provide largerembedding capacity and obtain stronger robustness againstMP3 compression and AWGN attacks The imperceptibilityis evaluated by using the ODG standardThe closer the ODGvalue to 0 the better the imperceptibility For the table it isnoted that the imperceptibility of the proposed method isbetter except for the clips track 35 and track 66 The reason isthat119864max values of track 35 and track 66 are bigger As a resultthresholds 119879 and 119866 are also larger and more embeddingdistortion will be caused
5 Conclusions
In this paper we proposed a robust and reversible audiowatermarking method by shifting the histogram of thestatistical feature values in time domainThe statistical featureis calculated as the sum of the prediction errors in a frameSince the audio clip has a larger number of samples andeach frame can hold enough elements the statistical featureis robust to those common signal processing operationsConsidering that the distribution of the statistical featurevalues may be distorted to some extent three extractionmethods and the majority voting system are designed forthe watermark detection Experimental results have shownthat thousands of bits can be reversibly embedded and thewatermark bits can resist MP3 compression of 64 kbps andadditive noise of 35 dB Comparingwith the existing excellentmethod the proposed method can embed more watermarkbits and achieve stronger robustness
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This work was supported by NSFC (no 61272414) and OpenResearch Fund from the State Key Laboratory of InformationSecurity (no 2016-MS-07)
References
[1] Y Xiang I Natgunanathan S GuoW Zhou and S NahavandildquoPatchwork-based audio watermarking method robust to de-synchronization attacksrdquo IEEE Transactions on Audio Speechand Language Processing vol 22 no 9 pp 1413ndash1423 2014
[2] J Fridrich M Goljan and R Du ldquoLossless data embeddingmdashnew paradigm in digital watermarkingrdquo EURASIP Journal onAdvances in Signal Processing vol 2002 no 2 pp 185ndash196 2002
[3] Y Q Shi Z Ni D Zou C Liang and G Xuan ldquoLosslessdata hiding fundamentals algorithms and applicationsrdquo inProceeding of the IEEE International Symposium on Circuits andSystems vol 2 pp 313ndash336 2004
[4] S Lee CD Yoo andTKalker ldquoReversible imagewatermarkingbased on integer-to-integer wavelet transformrdquo IEEE Transac-tions on Information Forensics and Security vol 2 no 3 pp 321ndash330 2010
[5] X Li B Yang and T Zeng ldquoEfficient reversible watermarkingbased on adaptive prediction-error expansion and pixel selec-tionrdquo IEEE Transactions on Image Processing vol 20 no 12 pp3524ndash3533 2000
[6] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hidingrdquo in Proceeding of the IEEE Interna-tional Conference onMultimedia and Expo (ICME rsquo2004) vol 3pp 2199ndash2202 Taipei Taiwan June 2004
[7] C De Vleeschouwer J Delaigle and B Macq ldquoCircularinterpretation of histogram for reversible watermarkingrdquo inProceeding of the IEEE 4th Workshop on Multimedia SignalProcessing pp 345ndash350 Cannes France 2001
[8] C De Vleeschouwer J F Delaigle and B Macq ldquoCircularinterpretation of bijective transformations in lossless water-marking for media asset managementrdquo IEEE Transactions onMultimedia vol 5 no 1 pp 97ndash105 2003
[9] D Zou YQ Shi ZNi andW Su ldquoA semi-fragile lossless digitalwatermarking scheme based on integer wavelet transformrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 10 pp 1294ndash1300 2006
[10] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hiding designed for semi-fragile imageauthenticationrdquo IEEE Transactions on Circuits and Systems forVideo Technology vol 18 no 4 pp 890ndash896 2008
[11] L An X Gao C Deng and F Ji ldquoRobust lossless data hidingAnalysis and evaluationrdquo in Procceding of the InternationalConference on High Performance Computing and Simulation(HPCS rsquo10) pp 512ndash516 July 2010
[12] X-T Zeng L-D Ping and X-Z Pan ldquoA lossless robust datahiding schemerdquo Pattern Recognition vol 43 no 4 pp 1656ndash1667 2010
[13] L An X Gao X Li D Tao C Deng and J Li ldquoRobustreversible watermarking via clustering and enhanced pixel-wisemaskingrdquo IEEE Transactions on Image Processing vol 21 no 8pp 3598ndash3611 2012
[14] R Thabit and B E Khoo ldquoCapacity improved robust losslessimage watermarkingrdquo IET Image Processing vol 8 no 11 pp662ndash670 2014
[15] S Xiang and Y Wang ldquoDistortion-free robust reversible water-marking by modifying and recording iwt means of imageblocksrdquo in Proceeding of the 14th International Workshop(IWDW rsquo15) Tokyo Japan October 2015
[16] D Coltuc and J Chassery ldquoDistortion-free robust watermark-ing a case studyrdquo in Security Steganography andWatermarkingof Multimedia Contents vol 6505 of Proceedings of SPIE pp588ndash595 San Jose Calif USA 2007
[17] A Nishimura ldquoReversible and robust audio watermarkingbased on spread spectrum and amplitude expansionrdquo in Inter-national Workshop on Digital Watermarking (IWDW rsquo14) vol9023 of Lecture Notes in Computer Science pp 215ndash229
10 Advances in Multimedia
[18] Massachusetts Institute of Technology (MIT) Audio Databasehttpsoundmediamitedumediaphp
[19] EBUCommittee Sound quality assessmentmaterial recordingsfor subjective tests httpstechebuchpublicationssqamcd
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal of
Volume 201
Submit your manuscripts athttpswwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 201
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
8 Advances in Multimedia
Table3Com
paris
onwith
metho
din
[17]
Track
Metho
din
[17]
Prop
osed
metho
dPayload
ODG
MP3
AWGN
119878119879
119866Payload
ODG
MP3
MP3
AWGN
(bits)
(128
Kbps)
(36d
B)(bits)
(80K
bps)
(64K
bps)
(36d
B)Track32
216
minus245
00
150
1500
3000
1000
minus137
002
0Track35
216
minus131
012
150
3500
5000
1000
minus158
008
0Track65
216
minus121
06
150
302000
1000
minus023
001
12
Track66
216
minus024
06
9660
0060
001000
minus105
03
12
0Track69
216
minus031
14
300
400
2000
1000
minus023
00
0
Advances in Multimedia 9
MP3 compression drops so the effect of the frame length 119878on the robustness against MP3 compression is unstable Theinfluence on AWGN is little
For fair comparison with the method in [17] we use thesame host signals (tracks 32 35 65 66 and 69) downloadedfrom sound quality assessment material (SQAM) collection[19] Table 3 shows the robustness testing results against MP3compression and additive noise (AWGN) operations We canobserve that the method in [17] can carry 216 bits and resisttheMP3 compression at 128Kbpswhile the proposedmethodcan resist the MP3 compression at 64Kbps with 1000 bitsembedded In addition in our method the BER under theAWGN of 35 dB is less than that of the method in [17]In other words the proposed method can provide largerembedding capacity and obtain stronger robustness againstMP3 compression and AWGN attacks The imperceptibilityis evaluated by using the ODG standardThe closer the ODGvalue to 0 the better the imperceptibility For the table it isnoted that the imperceptibility of the proposed method isbetter except for the clips track 35 and track 66 The reason isthat119864max values of track 35 and track 66 are bigger As a resultthresholds 119879 and 119866 are also larger and more embeddingdistortion will be caused
5 Conclusions
In this paper we proposed a robust and reversible audiowatermarking method by shifting the histogram of thestatistical feature values in time domainThe statistical featureis calculated as the sum of the prediction errors in a frameSince the audio clip has a larger number of samples andeach frame can hold enough elements the statistical featureis robust to those common signal processing operationsConsidering that the distribution of the statistical featurevalues may be distorted to some extent three extractionmethods and the majority voting system are designed forthe watermark detection Experimental results have shownthat thousands of bits can be reversibly embedded and thewatermark bits can resist MP3 compression of 64 kbps andadditive noise of 35 dB Comparingwith the existing excellentmethod the proposed method can embed more watermarkbits and achieve stronger robustness
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This work was supported by NSFC (no 61272414) and OpenResearch Fund from the State Key Laboratory of InformationSecurity (no 2016-MS-07)
References
[1] Y Xiang I Natgunanathan S GuoW Zhou and S NahavandildquoPatchwork-based audio watermarking method robust to de-synchronization attacksrdquo IEEE Transactions on Audio Speechand Language Processing vol 22 no 9 pp 1413ndash1423 2014
[2] J Fridrich M Goljan and R Du ldquoLossless data embeddingmdashnew paradigm in digital watermarkingrdquo EURASIP Journal onAdvances in Signal Processing vol 2002 no 2 pp 185ndash196 2002
[3] Y Q Shi Z Ni D Zou C Liang and G Xuan ldquoLosslessdata hiding fundamentals algorithms and applicationsrdquo inProceeding of the IEEE International Symposium on Circuits andSystems vol 2 pp 313ndash336 2004
[4] S Lee CD Yoo andTKalker ldquoReversible imagewatermarkingbased on integer-to-integer wavelet transformrdquo IEEE Transac-tions on Information Forensics and Security vol 2 no 3 pp 321ndash330 2010
[5] X Li B Yang and T Zeng ldquoEfficient reversible watermarkingbased on adaptive prediction-error expansion and pixel selec-tionrdquo IEEE Transactions on Image Processing vol 20 no 12 pp3524ndash3533 2000
[6] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hidingrdquo in Proceeding of the IEEE Interna-tional Conference onMultimedia and Expo (ICME rsquo2004) vol 3pp 2199ndash2202 Taipei Taiwan June 2004
[7] C De Vleeschouwer J Delaigle and B Macq ldquoCircularinterpretation of histogram for reversible watermarkingrdquo inProceeding of the IEEE 4th Workshop on Multimedia SignalProcessing pp 345ndash350 Cannes France 2001
[8] C De Vleeschouwer J F Delaigle and B Macq ldquoCircularinterpretation of bijective transformations in lossless water-marking for media asset managementrdquo IEEE Transactions onMultimedia vol 5 no 1 pp 97ndash105 2003
[9] D Zou YQ Shi ZNi andW Su ldquoA semi-fragile lossless digitalwatermarking scheme based on integer wavelet transformrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 10 pp 1294ndash1300 2006
[10] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hiding designed for semi-fragile imageauthenticationrdquo IEEE Transactions on Circuits and Systems forVideo Technology vol 18 no 4 pp 890ndash896 2008
[11] L An X Gao C Deng and F Ji ldquoRobust lossless data hidingAnalysis and evaluationrdquo in Procceding of the InternationalConference on High Performance Computing and Simulation(HPCS rsquo10) pp 512ndash516 July 2010
[12] X-T Zeng L-D Ping and X-Z Pan ldquoA lossless robust datahiding schemerdquo Pattern Recognition vol 43 no 4 pp 1656ndash1667 2010
[13] L An X Gao X Li D Tao C Deng and J Li ldquoRobustreversible watermarking via clustering and enhanced pixel-wisemaskingrdquo IEEE Transactions on Image Processing vol 21 no 8pp 3598ndash3611 2012
[14] R Thabit and B E Khoo ldquoCapacity improved robust losslessimage watermarkingrdquo IET Image Processing vol 8 no 11 pp662ndash670 2014
[15] S Xiang and Y Wang ldquoDistortion-free robust reversible water-marking by modifying and recording iwt means of imageblocksrdquo in Proceeding of the 14th International Workshop(IWDW rsquo15) Tokyo Japan October 2015
[16] D Coltuc and J Chassery ldquoDistortion-free robust watermark-ing a case studyrdquo in Security Steganography andWatermarkingof Multimedia Contents vol 6505 of Proceedings of SPIE pp588ndash595 San Jose Calif USA 2007
[17] A Nishimura ldquoReversible and robust audio watermarkingbased on spread spectrum and amplitude expansionrdquo in Inter-national Workshop on Digital Watermarking (IWDW rsquo14) vol9023 of Lecture Notes in Computer Science pp 215ndash229
10 Advances in Multimedia
[18] Massachusetts Institute of Technology (MIT) Audio Databasehttpsoundmediamitedumediaphp
[19] EBUCommittee Sound quality assessmentmaterial recordingsfor subjective tests httpstechebuchpublicationssqamcd
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal of
Volume 201
Submit your manuscripts athttpswwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 201
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
Advances in Multimedia 9
MP3 compression drops so the effect of the frame length 119878on the robustness against MP3 compression is unstable Theinfluence on AWGN is little
For fair comparison with the method in [17] we use thesame host signals (tracks 32 35 65 66 and 69) downloadedfrom sound quality assessment material (SQAM) collection[19] Table 3 shows the robustness testing results against MP3compression and additive noise (AWGN) operations We canobserve that the method in [17] can carry 216 bits and resisttheMP3 compression at 128Kbpswhile the proposedmethodcan resist the MP3 compression at 64Kbps with 1000 bitsembedded In addition in our method the BER under theAWGN of 35 dB is less than that of the method in [17]In other words the proposed method can provide largerembedding capacity and obtain stronger robustness againstMP3 compression and AWGN attacks The imperceptibilityis evaluated by using the ODG standardThe closer the ODGvalue to 0 the better the imperceptibility For the table it isnoted that the imperceptibility of the proposed method isbetter except for the clips track 35 and track 66 The reason isthat119864max values of track 35 and track 66 are bigger As a resultthresholds 119879 and 119866 are also larger and more embeddingdistortion will be caused
5 Conclusions
In this paper we proposed a robust and reversible audiowatermarking method by shifting the histogram of thestatistical feature values in time domainThe statistical featureis calculated as the sum of the prediction errors in a frameSince the audio clip has a larger number of samples andeach frame can hold enough elements the statistical featureis robust to those common signal processing operationsConsidering that the distribution of the statistical featurevalues may be distorted to some extent three extractionmethods and the majority voting system are designed forthe watermark detection Experimental results have shownthat thousands of bits can be reversibly embedded and thewatermark bits can resist MP3 compression of 64 kbps andadditive noise of 35 dB Comparingwith the existing excellentmethod the proposed method can embed more watermarkbits and achieve stronger robustness
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
Acknowledgments
This work was supported by NSFC (no 61272414) and OpenResearch Fund from the State Key Laboratory of InformationSecurity (no 2016-MS-07)
References
[1] Y Xiang I Natgunanathan S GuoW Zhou and S NahavandildquoPatchwork-based audio watermarking method robust to de-synchronization attacksrdquo IEEE Transactions on Audio Speechand Language Processing vol 22 no 9 pp 1413ndash1423 2014
[2] J Fridrich M Goljan and R Du ldquoLossless data embeddingmdashnew paradigm in digital watermarkingrdquo EURASIP Journal onAdvances in Signal Processing vol 2002 no 2 pp 185ndash196 2002
[3] Y Q Shi Z Ni D Zou C Liang and G Xuan ldquoLosslessdata hiding fundamentals algorithms and applicationsrdquo inProceeding of the IEEE International Symposium on Circuits andSystems vol 2 pp 313ndash336 2004
[4] S Lee CD Yoo andTKalker ldquoReversible imagewatermarkingbased on integer-to-integer wavelet transformrdquo IEEE Transac-tions on Information Forensics and Security vol 2 no 3 pp 321ndash330 2010
[5] X Li B Yang and T Zeng ldquoEfficient reversible watermarkingbased on adaptive prediction-error expansion and pixel selec-tionrdquo IEEE Transactions on Image Processing vol 20 no 12 pp3524ndash3533 2000
[6] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hidingrdquo in Proceeding of the IEEE Interna-tional Conference onMultimedia and Expo (ICME rsquo2004) vol 3pp 2199ndash2202 Taipei Taiwan June 2004
[7] C De Vleeschouwer J Delaigle and B Macq ldquoCircularinterpretation of histogram for reversible watermarkingrdquo inProceeding of the IEEE 4th Workshop on Multimedia SignalProcessing pp 345ndash350 Cannes France 2001
[8] C De Vleeschouwer J F Delaigle and B Macq ldquoCircularinterpretation of bijective transformations in lossless water-marking for media asset managementrdquo IEEE Transactions onMultimedia vol 5 no 1 pp 97ndash105 2003
[9] D Zou YQ Shi ZNi andW Su ldquoA semi-fragile lossless digitalwatermarking scheme based on integer wavelet transformrdquoIEEE Transactions on Circuits and Systems for Video Technologyvol 16 no 10 pp 1294ndash1300 2006
[10] Z Ni Y Q Shi N Ansari W Su Q Sun and X Lin ldquoRobustlossless image data hiding designed for semi-fragile imageauthenticationrdquo IEEE Transactions on Circuits and Systems forVideo Technology vol 18 no 4 pp 890ndash896 2008
[11] L An X Gao C Deng and F Ji ldquoRobust lossless data hidingAnalysis and evaluationrdquo in Procceding of the InternationalConference on High Performance Computing and Simulation(HPCS rsquo10) pp 512ndash516 July 2010
[12] X-T Zeng L-D Ping and X-Z Pan ldquoA lossless robust datahiding schemerdquo Pattern Recognition vol 43 no 4 pp 1656ndash1667 2010
[13] L An X Gao X Li D Tao C Deng and J Li ldquoRobustreversible watermarking via clustering and enhanced pixel-wisemaskingrdquo IEEE Transactions on Image Processing vol 21 no 8pp 3598ndash3611 2012
[14] R Thabit and B E Khoo ldquoCapacity improved robust losslessimage watermarkingrdquo IET Image Processing vol 8 no 11 pp662ndash670 2014
[15] S Xiang and Y Wang ldquoDistortion-free robust reversible water-marking by modifying and recording iwt means of imageblocksrdquo in Proceeding of the 14th International Workshop(IWDW rsquo15) Tokyo Japan October 2015
[16] D Coltuc and J Chassery ldquoDistortion-free robust watermark-ing a case studyrdquo in Security Steganography andWatermarkingof Multimedia Contents vol 6505 of Proceedings of SPIE pp588ndash595 San Jose Calif USA 2007
[17] A Nishimura ldquoReversible and robust audio watermarkingbased on spread spectrum and amplitude expansionrdquo in Inter-national Workshop on Digital Watermarking (IWDW rsquo14) vol9023 of Lecture Notes in Computer Science pp 215ndash229
10 Advances in Multimedia
[18] Massachusetts Institute of Technology (MIT) Audio Databasehttpsoundmediamitedumediaphp
[19] EBUCommittee Sound quality assessmentmaterial recordingsfor subjective tests httpstechebuchpublicationssqamcd
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal of
Volume 201
Submit your manuscripts athttpswwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 201
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
10 Advances in Multimedia
[18] Massachusetts Institute of Technology (MIT) Audio Databasehttpsoundmediamitedumediaphp
[19] EBUCommittee Sound quality assessmentmaterial recordingsfor subjective tests httpstechebuchpublicationssqamcd
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal of
Volume 201
Submit your manuscripts athttpswwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 201
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of
RoboticsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Active and Passive Electronic Components
Control Scienceand Engineering
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
RotatingMachinery
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporation httpwwwhindawicom
Journal of
Volume 201
Submit your manuscripts athttpswwwhindawicom
VLSI Design
Hindawi Publishing Corporationhttpwwwhindawicom Volume 201
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Shock and Vibration
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Civil EngineeringAdvances in
Acoustics and VibrationAdvances in
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Electrical and Computer Engineering
Journal of
Advances inOptoElectronics
Hindawi Publishing Corporation httpwwwhindawicom
Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
SensorsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Navigation and Observation
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
DistributedSensor Networks
International Journal of