secure and efficient drm watermark algorithm of forensics

RESEARCH Open Access

Secure and efficient DRM watermarkalgorithm of forensics in mobile internetMa Zhaofeng1* and Jiang Ming2

Abstract

With the development of mobile Internet technology, the characteristic of easy editing, transmitting, and forging thedigital media bring great challenges in authenticity of multimedia. Due to that, the focus on the digital forensics andidentification, such as source identification, content authentication, and information integrity, have become an importantresearch content in the field network security, forensic science and other fields. To solve this problem, in this paper, weproposed an image digital rights management scheme for mobile Internet forensics based on watermark with high-levelsecurity and traceability. In this scheme, we embed some user-identity-related and device-related information of mobilephones as a source identification of snapped photos and stored pictures; once the forwarded image data from mobilephones are misused such as spreading the image data on Internet without authorization, especially the confidentialimage data, we can trace the misuse responsibility by extracting the mobile phone data embedded in the mobilephotos. Finally, we evaluate the proposed security and efficiency of the DRM watermark scheme for mobile Internet,a large amount of experiments manifest the proposed watermark scheme is robust, secure, and efficient for theprotection and misuse tracing of mobile image data.

Keywords: Digital rights management, Watermark, Forensics

1 IntroductionDigital forensics is the technology of use of scientificallyderived and proven methods toward the preservation,collection, validation, identification, interpretation, docu-mentation, and presentation of digital evidence derivedfrom digital sources for the purpose of facilitating orfurthering there construction of events found to becriminal or helping to anticipate unauthorized actionsshown to be disruptive to planned operations. Themeaning of the term digital evidence refers to any infor-mation that can provide evidence, which contains storedor digitally transmitted data, and they can be used incourt as evidence.In the past few years, much more digital evidences

have been used in cases, such as documents, e-mail,digital photos, message history, historical records data-base of browser, computer memory data, GPS trackingdata, and digital images or sound files.

With the rapid development of multimedia, especiallythe popularity of the mobile and the convenience of themobile Internet, much more image data produced frommobiles are much easier to spread in the circulation ofthe mobile Internet. According to statistics, by the endof June 2017, the total number of mobile phone users inChina is 1 billion 360 million, and the Internet users ofmobile phones break through 1 billion 100 million. Theconfiguration of mobile camera is more advanced, andpeople use the portable mobile phone much moreconveniently to get high-quality photos of a scene in life.The close connection between mobile phones and peo-ple’s lives makes much more cases based on the evidenceof mobile data happened in the judicial practice. Digitalimage forensic techniques are the technologies andmethodologies for judging the origin, integrity, andauthenticity of digital images, by analyzing, identifying,and authenticating the steganography, forgery, and tam-pering of digital images. There are two main research di-rections in digital image forensic techniques [1]: imagesource detection and tamper detection [2].For mobile image source detection, the lens and im-

aging sensors of different brands and models of mobiles

* Correspondence: [email protected] Security Center of Beijing University of Posts andTelecommunications, Beijing 100876, ChinaFull list of author information is available at the end of the article

EURASIP Journal on Imageand Video Processing

© The Author(s). 2018 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link tothe Creative Commons license, and indicate if changes were made.

Zhaofeng and Ming EURASIP Journal on Image and Video Processing (2018) 2018:69 https://doi.org/10.1186/s13640-018-0307-5

http://crossmark.crossref.org/dialog/?doi=10.1186/s13640-018-0307-5&domain=pdf

http://orcid.org/0000-0001-7875-4169

mailto:[email protected]

http://creativecommons.org/licenses/by/4.0/

are different, with which the process of digital imagesare also different, including compression and storage.Therefore, we extract and analyze the small differ-ences, such as image style and quality, to identify thesource of the digital mobile images. There are threemain types of detection technologies for imagesources forensics: (1) Mobile digital image sources aretaken as the evidence based on mobile device type:we use a classifier to classify mobile image sources byextracting statistical characteristics, such as imagequality, color, lens radial distortion, and wavelet coef-ficients. Levent Özparlak proposed statistical imagemodels for wavelet-based transforms and comparedtheir relative merits within the context of digitalimage forensics [3]. Farid described a series of psy-chophysical experiments that used images of varyingresolutions, JPEG compression, and colors to explorethe ability of observers to distinguish computer-gener-ated from photographic images of people [4]. The re-searchers also put forward some effective features todistinguish and identify the digital mobile and CGimages, such as CFA interpolation periodicity [5], vis-ual characteristics [6], texture feature [7, 8], whitebalance [9], and local binary pattern(LBP) [10]. Formore different types of mobile devices, Orozco usedcolor features and quality features [11]. (2) Mobiledigital image sources are taken as the evidence bymobile device model: the same tomography hardwareand the same image processing algorithm is used inmobiles with the same device model, so we use thecharacteristics of hardware device and image process-ing algorithm to be the important foundation for mo-bile image forensics. Chen [12] estimated theinterpolation coefficient of CFA and selected a suit-able classifier to achieve higher accuracy of sourceidentification. White balance [13] is an importantimage post-processing algorithm in mobile and cam-era imaging system, and its parameter estimation isalso applied to identify the source of digital image.Another part of research work takes the entire mobileimage acquisition equipment as a whole and expectsto build a whole model to describe the difference inthe mobile image equipment from different anglesand realize the source forensics. Goyal [14] evaluatesthe effectiveness of image quality measures (IQM) foridentifying the source mobile phone from the imagesor videos captured by that mobile phone. A newsingle-image feature for mobile identification is pro-posed based on a local binary pattern of extractededges form the input mobile image [15]. Thai pro-posed a statistical approach for the camera modelidentification problem, which is based on the hetero-scedastic noise model for describing a natural rawimage [16]. Unlike the preceding methods, Luan takes

the source forensics as a clustering problem in un-supervised learning. To avoid using any prior know-ledge in practical scenarios, Luan proposed agraph-based approach to classify the source cell-phones[17]. (3) Mobile digital image sources are taken as the evi-dence by mobile device individual characteristics: we candetermine the mobile image source by correlation detec-tion between the pattern noise and abnormal pixelscaused by inherent defects of the imaging device. Fridrich[18] divides sensor mode noise into inherent mode noiseand optical response non-uniformity noise, obtains themode noise of digital camera by filtering and statistical dif-ference, and realizes the forensics of camera individualsources of digital images using correlation detection, hy-pothesis testing, and other methods. But besides that all,sensor mode noise are also used for device individualsource forensics in mobile [19], portable digital video cam-era [20], and scanner [21].For mobile image tamper detection, the tamper and

the collector of the mobile images do against with eachother. The tamper is to make false images of truth asfast as you can, while the collector is to try the best tofind evidence of the image being tampered with. Al-though it is difficult to be easily discovered, the tam-pered image causes more or less damage to the inherentcontinuity of the natural image during the capturingprocess and storing process. Therefore, the collectorsuse some statistical characteristics of natural images todetect the types of image tampered and display the de-tailed location of tampering. There are also three maintypes of detection technologies for image tamper foren-sics. (1) The first method is based on resampling, suchas copy-paste [22, 23] and splice-synthesis [24]. (2) Thesecond method is based on compression, such as dualcompression [25] and JPEG blocking artifact [26, 27]. (3)The third method is based on imaging device features,such as image statistical characteristics [28], patternnoise [29] and CFA interpolation [30–35].

2 Methods2.1 Watermark algorithm principle for source forensic andtamper detectionIn order to solve the problem of source detection andtamper detection of mobile image data, we proposed anew watermark-based mobile image data securityscheme to protect confidential image data in this paper.Robust watermark, such as mobile-identification-relatedand user-identification-related information, are used asin this scheme for source detection, while the features ofimage itself are used as semi-fragile watermark in thisscheme for traceability. Finally, for security and effi-ciency, we evaluated the proposed mobile image data se-curity scheme through several sets of experiments.

Zhaofeng and Ming EURASIP Journal on Image and Video Processing (2018) 2018:69 of 14

As the storage capacity of the mobile phone is limited,the image data is stored and transmitted in a compressedway. Considering JPEG picture compression format iswidely used in mobile phones, we use JPEG as an exampleof confidential image data in our proposed mobile imagedata security solution. In this scheme, robust watermarkfor source forensics is a binary image, which indicatessome text info related to mobile identification and useridentification. Semi-fragile watermark data for tamper de-tecting is also a binary image, which is generated by thehost image itself. In order to avoid the impact between thetwo digital watermarks, the two watermarking algorithmsshould be independent. Considering JPEG compression islossy, which may bring some errors to watermarks, thewatermark scheme should be robust to JPEG compressionat least. Discrete cosine transform (DCT) is reversible inJPEG, and the DCT coefficients have a certain degree ofindependence, so we choose to embed watermarks basedon the DCT coefficients in this scheme.

2.2 JPEG image compression and DCT transformJPEG compression usually divides the completely digitalimage into 8 × 8 blocks, based on DCT, and quantifiesthe luma samples of each block (see Fig. 1). The DCT oftwo-dimensional signal f(i, j), i = 0, 1, …M − 1; j = 0, 1, …N − 1) f (i, j), i = 0, 1, …M − 1; j = 0, 1, …N − 1) is definedas follows:

F u; vð Þ ¼ c uð Þc vð ÞXM−1

i¼0

XN−1

j¼0

f i; jð Þ cos 2iþ 1ð Þuπ2M

cos2 jþ 1ð Þvπ

2N

u ¼ 0; 1;…;M−1; v ¼ 0; 1;…;N−1: ð1Þ

The inverse IDCT is defined as follows:

f i; jð Þ ¼XM−1

u¼0

XN−1

v¼0

c uð Þc vð ÞF u; vð Þ cos 2iþ 1ð Þuπ2M

cos2 jþ 1ð Þvπ

2N

u ¼ 0; 1;…;M−1; v ¼ 0; 1;…;N−1: ð2Þ

c uð Þ ¼1ffiffiffiffiffiM

p u ¼ 0

2ffiffiffiffiffiM

p u≠0

8>><>>:

9>>=>>; c vð Þ ¼

1ffiffiffiffiN

p v ¼ 0

2ffiffiffiffiN

p v≠0

8>><>>:

9>>=>>;

2.3 JPEG quantizationJPEG encodes DC coefficient using differential pulsecode modulation and encodes AC coefficients in a zigzagsequence, as shown in Fig. 2a. Figure 2b is thequantization table for JPEG compression.

2.4 Masking model of HVSThe main idea of the adaptive watermarking methodbased on human visual mode is HVS that can help usembed watermarks in different regions with differentintensities adaptively, which means we focus on min-imizing watermark strength in regions of image sensi-tive to HVS and maximizing watermark strength inregions of image not sensitive to HVS. Texture mask-ing and illuminance masking indicate we shouldembed watermark intensively in the regions of imageswith high texture and brightness.Performing two-level wavelet transform on the image,

the quad tree structure is shown in Fig. 3. The model ofthe illuminance sensitivity is shown in Fig. 4.Supposing two-dimensional image is f(x, y) ∈

L2 (R2), two-dimensional wavelet decompositionimage is:

Aj f ¼ Ajþ1 f þ D2jþ1 f þ D3

jþ1 ð3Þ

Ajþ1 fP∞

m¼−∞

P∞n¼−∞ C jþ1ðm; nÞ∅ jþ1ðm; nÞDi

jþ1

¼P∞m¼−∞

P∞n¼−∞ Di

jþ1ðm; nÞ∅ijþ1ðm; nÞ i = 1, 2, 3 (4)

C jþ1 m; nð Þ ¼X∞k¼−∞

X∞l¼−∞

h k−2mð Þh l−2nð ÞC j k; lð Þ

D1jþ1 m; nð Þ ¼

X∞k¼−∞

X∞l¼−∞

h k−2mð Þh l−2nð ÞC j k; lð Þ

D2jþ1 m; nð Þ ¼

X∞k¼−∞

X∞l¼−∞

h k−2mð Þg l−2nð ÞC j k; lð Þ

D3jþ1 m; nð Þ ¼

X∞k¼−∞

X∞l¼−∞

h k−2mð Þg l−2nð ÞC j k; lð Þ

8>>>>>>>>>>>>><>>>>>>>>>>>>>:

ð5Þwhere h(∙) and g(∙) are the horizontal filter function and

Fig. 1 DCT transform


vertical filter function, respectively. j means the compos-ition level.Texture masking matrix is M ×N, shown as Ts .

Ts m; nð Þ ¼ δ LH1ð Þ þ δ HL1ð Þ þ δ HH1ð Þ ð6Þ

δ ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

1M � N

XM−1

m¼0

XN−1

n¼0

x m; nð Þ−μð Þ2vuut ð7Þ

μ ¼ 1M � N

XM−1

m¼0

XN−1

n¼0

x m; nð Þ2 ð8Þ

m ¼ 0; 1; 2;…;M−1; n ¼ 0; 1; 2;…;N−1:

The normalization of Ts is recorded as TM .Normalizing the LL2 coefficient matrix, illuminance

masking matrix is expressed as Ls. The size of matrixLL2 is M ×N,

L m; nð Þ ¼ f LL2 m; nð Þð Þ ð9Þ

f xð Þ ¼−

1180

xþ 3:5 0≤x < 90

−338

xþ 19219

90≤x < 128

2127

x−256127

128≤x < 255

8>>>><>>>>:

9>>>>=>>>>;

ð10Þ

m ¼ 0; 1; 2;…;M−1; n ¼ 0; 1; 2;…;N:

where f (∙) is a function of the model of Illuminance sen-sitivity, and normalization of Ls is recorded as LM .The masking matrix is as follows:

SS ¼ a∙LM þ b∙TM ð11Þ

where a = 1 − b. Normalization of SS is recorded as SM.Figure 5 reflects the much higher brightness part of

the image, the much more suitable for embeddingwatermark.

Fig. 2 a and b is the DTC quantization for JPEG compression

Fig. 3 The quadtree structure of DWT for Lena is shown in Fig. 3


JPEG standard is based on 8 × 8 blocks, and themasking of visual for block image is calculated out bythe average masking of pixels of block image, re-corded as Q.

2.5 The proposed mobile image data security schemebased on watermark2.5.1 Watermark scheme architecture for mobile internetforensicsWatermark scheme architecture for mobile internetforensics is shown in the following Fig. 6, which in-cludes the source forensic watermark algorithm,tamper tracing watermark algorithm, and tamper de-tecting algorithm.

2.5.2 Watermark embedding algorithm for source forensicsIn the watermark embedding process, we choose oneM ×N binary image as watermark W. W = {W(i, j)| 0 ≤ i<M, 0 ≤ j <N}, and W(i, j) ∈ {0, 1}. We scramble the

binary image for security and then convert the imageinto a one-dimensional watermark sequence, namely W= {wi}, i = 1, 2, …, C; C =M ×N, wi = 0 or 1. The hostimage is divided into 8 × 8 blocks, named Xi, and xi(m,n) is the pixel value in (m, n) of Xi. The sequence ofDCT coefficients of one block is recorded as Ci(j), (j = 0,1, 2,…63) in a zigzag scan form. i is the number ofblocks. We choose a continuum of mid-frequency coeffi-cients to embed the watermark, such as

Ci k−2ð Þ;Ci k−1ð Þ;Ci kð Þ;Ci k þ 1ð Þ;Ci k þ 2ð Þ

k ¼ 2; 3;…; 61

The processing of coefficient Ci(k) balances the trans-parency and robustness of watermark. The specificmethods are as follows:if wi = 0, then we can get

Fig. 4 Model of illuminance sensitivity

Fig. 5 Texture image of Lena. a = 0.1. a = 0.3. a = 0.5. a = 0.7. a = 0.9


Ci kð Þ0 ¼ MIN Ci k−2ð Þ;Ci k−1ð Þ;Ci kð Þ;Ci k þ 1ð Þ;Ci k þ 2ð Þð Þ−Qi

ð12Þelse if wi = 1, then we can get

Ci kð Þ0 ¼ MAX Ci k−2ð Þ;Ci k−1ð Þ;Ci kð Þ;Ci k þ 1ð Þ;Ci k þ 2ð Þð Þ þ Qi

ð13ÞCi(k)

′is the modified coefficient, MIN(.) denotes theminimum value function, MAX(.) denotes the maximumfunction. and Qi can control the watermark strength. Qi

is defined in Fig. 7.Five consecutive DCT coefficients are being as host

sequences, and how to select the coefficients is thekey to the algorithm, which directly affects the per-formance of the watermark. As shown in Fig. 1, dif-ferent Ci(k) means different impacts on block images.In order to balance the transparency and robustnessof the watermark in this algorithm, we select

medium- and low-frequency coefficient watermark toembed. Experimental analysis of different Ci(k) andPSNR has been carried out in Fig. 8. The host imageis the standard image Lena, and the watermark isfrom a pseudo-random sequence.In the graph, the horizontal ordinate is the location

of the watermark coefficient that is the k of Ci(k).The vertical ordinate is the PSNR of watermarkedJPEG image. Overall, with the offset of the embeddedpoint, embedding a watermark in low frequency maybring low imperceptibility, while embedding water-mark in high frequency may bring better visual qual-ity of the watermarked image. The maximum valueappears in the curve line when k is 12, in which thePSNR is 42.78 dB. The main reasons may be asfollows:

1. From Fig. 1, we can see the change of mediumfrequency Ci(12) impact the block image evenly.

Fig. 6 Flow of watermark embed scheme

Fig. 7 The masking characteristic of Image


2. When k is 12, the modification of the coefficient isin relation to five consecutive coefficients, such asCi(10), Ci(11), Ci(12), Ci(13), and Ci(14), which arein the same layer in the zigzag scanning line andvery even.

3. As show in Fig. 2b, the quantization step is 8 whenthe k is 12, which is smaller than the steps around, sothe quantization errors may be smaller and bringhigh PSNR of block images.

From the above analysis, we select Ci(12) (k = 12) asthe watermark embedding point.From the formula 11, we can get Qi that is related to

the texture of the image. The much bigger Qi means thehigher texture in the blocks of the image and the easierto embed watermarks. The texture image of Lena usingdifferent b is as shown in Fig. 5.

2.5.3 Watermark embedding algorithm for tamper tracingExcept for source forensics, the scheme also needs toembed another watermark for tamper tracing of the mo-bile images on the Internet. Because the host image willbe saved in JPEG, the semi-fragile watermark for tampertracing should be robust to JPEG compression. To avoidattacking the watermark for source forensics embeddedin DCT AC mid-frequency coefficients, the watermarksfor tamper tracing are chosen to be embedded in DCTDC coefficients. For blind watermarking algorithm, amethod of coefficient quantization will be used. Theprocedures are as follows:

Step 1: The algorithm creates a two-valued pseudo-random sequence using the length and width of thehost image.

S ¼ Random M;N ; kð Þ ð14Þ

M and N represent the length and width of the image,respectively. k is a key for random function. S is atwo-value sequence with only 0 and 1 (si = 0 or 1), thelength of which is the number of block images. Random(.) is the function for pseudo-random.

Step 2: And then,

Ci 0ð Þ0 ¼

if si ¼ 0

Q Ci 0ð Þð Þ � γ þ γ2

l mþ mod Ci 0ð Þ; 2ð Þ � γ

� �þ 1:5� γ

else if si ¼ 1

Q Ci 0ð Þð Þ � γ þ γ2

l mþ mod Ci 0ð Þ þ 1; 2ð Þ � γ

� �þ 1:5� γ

8>>>>>>>><>>>>>>>>:

Q Ci 0ð Þð Þ ¼ Ci 0ð Þ−1γ

� �ð15Þ

In the formula above, Ci(0)′ is the modified coefficient

after watermarking algorithm, Ci(0) is the DC coefficientof every 8 × 8 block image, γ describes the quantizationsteps, iis the sequence number corresponding to the ithblock of the image. In order to detect the tamper regionin the entire image, all blocks of images should beembed the watermark using the formula 15.

Fig. 8 PSNR for different k


2.5.4 Tamper detecting process of watermark schemeAfter the source forensic algorithm and tamper tra-cing algorithm, the DCT coefficients are compressedinto JPEG images after quantization and coding. Inthe scheme, the forensic algorithm is robust and thetamper tracing algorithm is semi-fragile. Lossy JPEGcompression may cause some errors in watermarkextracting, which should be verified that has no ef-fects on the tamper tracing detecting. The low com-pression rate causes the size of the watermarkedJPEG images to be too large especially for mobilestorage; meanwhile, the high compression rate maycause some errors in tamper detection. Therefore, inthe scheme, we carry out one tamper detectingprocess to choose the most suitable quality factor forJPEG compression, which is very important for thesize of JPEG saving images. The tamper detectingprocess is described as follows:

Step 1: Store mobile JPEG image watermarked withquality factor Q = 100. The PSNR of thewatermarked JPEG image is very high, but the sizeof image is too large. The choice of the qualityfactor is very important for the watermarkperformance, so the performance evaluation will bego on next;

Step 2: Conduct performance evaluation onwatermarked JPEG image. Performance evaluationmainly considers three aspects: tampering, the CIR(formula 20), and the bit error rate of the robustwatermark for source forensic. If there is notamper, the CIR and bit error are in a certainthreshold range, go to step 3, or else go to step 4;the specific tamper detect methods are as follows:

Read watermarked JPEG image, based on DCT, andquantify the luma samples of each block:

1. : We can get si like the method as formula 14.2. : And then,

si0 ¼

0 mod Q Ci 0ð Þð Þ; 2ð Þ ¼ 0

1 otherwise

8<:

Q Ci 0ð Þð Þ ¼ Ci 0ð Þ−1γ

� �ð16Þ

if ðs0i ¼¼ siÞ , there is no tamper in block image, elsethe block is tampered.Ci(0) is the DC coefficient in the ith 8 × 8 block image,

F(.) indicates the rounding down function, and γ de-scribes half the quantization step. If the tamper block is

needed to be labeled, we can write the block imageblack.

Step 3: Since there is no tampering, we can save thewatermarked JPEG image with the new compressionquality Q =Q − 1. And then, the size of the imagemay be smaller, which is very important for filestoring in a mobile. After that, we also should verifythe tamper detection, go to step 2;

Step 4: Once any block image is tampered, whichshows that the quality factor of JPEG compressionis too small and the lossy compression affects thetamper detection of the watermark, the qualityfactor needs to be improved. Store the watermarkedJPEG image with the quality factor Q =Q + 1, thenthe whole watermark embed scheme is complete.

2.5.5 Watermark extracting algorithm

2.5.5.1 Watermark extracting algorithm for sourceforensic In order to take the evidence for images fromthe mobile Internet, we should extract the forensicwatermark in the suspicious target image. After gettingthe JPEG images from a mobile, we still need to dividethe JPEG image into 8 × 8 blocks, based on DCT, andquantify the luma samples of each block, and we alsoget the DCT coefficients Ci(j), (j = 0, 1, 2,…63) the sameas the watermark embedding process, i is the sequencenumber corresponding to the ith block of the image. Weselect the mid-frequency coefficients to extract wa-termarks, such as Ci(k − 2), Ci(k − 1), Ci(k), Ci(k + 1),Ci(k + 2) corresponding with the embed process.Extracting method is as follows:

if Ci kð Þ >�Ci k−2ð Þ þ Ci k−1ð Þ;þCi kð Þ þ Ci kþ 1ð Þ

þ Ci kþ 2ð Þ�=5;wi ¼ 1:

elsewi = 0.(17)

where wi is the ith watermark bit. At last,anti-scramble the watermark bit to get the watermarkimage.

2.5.5.2 Watermark extracting algorithm for tamperdetecting The process of tamper detection is describedas step 2 in Section 2.5.4. If one block image is tam-pered, mark it with black.

3 Results and discussions3.1 Experiments and evaluations of the DRM watermarkalgorithmIn this scheme, we designed watermark-based effectivedigital rights management (DRM) algorithm for mobile


Internet forensics by embedding watermarks based onDCT, in which the robust watermark is for forensicswhile semi-fragile watermark is for tamper tracing. Inour scheme, we constructed a watermark in a figureformat rather than in a text format; thus, even if theimage data is attacked much more, the watermarkcan still be recognized the figured watermark and canparse the user-related and hardware-related informa-tion and then can take the evidence by the figuredwatermark.In order to evaluate the scheme, we carry out various

experiments. In terms of watermark invisibility, PSNR isused for evaluating image quality objectively, which isdefined as:

PSNR ¼ 10 log10O−1ð Þ2MSE

!dB ð18Þ

where O − 1 represents the maximum value of the ori-ginal image pixels. MSE is the mean squared errors,given by

MSE ¼ 1M � N

XM−1

i¼0

XN−1

j¼0

W i; jð Þ−W i; jð Þ� �2 ð19Þ

In this experiment, some photos taken by a mobilephone with the size of 3120 × 4160 is used. A binary logoof size 250 × 250 is used as a watermark, which describessome info for the mobile phone, such as phone no., time,location and IMEI. Figure 9 shows the original image,binary watermark, and the corresponding watermarked

Fig. 9 Watermark embedding and extracting

Table 1 Different quality factors for stadium image

Quality factor CIR (%) PSNR (dB) Bit error (%) Is tampered?

100 169.69 39.00 0 NO

95 42.25 39.03 0 NO

90 11.26 38.64 0 NO

85 − 10.7 38.99 0.0001 NO

80 − 18.34 38.44 0.0004 NO

75 − 23.57 37.80 0.0010 NO

70 − 28.26 37.36 0.0017 NO


image. The PSNR of the watermarked image is 38.87 dBwhile the quality factor is 93.After the watermark is embedded, the watermarked

JPEG file storage capacity may be increased. To con-trol storage capacity, the watermarked JPEG image

should be compressed and stored at a low quality fac-tor. We use the index CIR (capacity increase ratio) asthe percentage of image capacity increase-ratio be-tween the host JPEG image and the watermarkedJPEG image.

Fig. 10 Watermark attacking and detecting


Fig. 11 Different mobile pictures with watermarks


CIR ¼ Capacity watermarkedð Þ−Capacity originalð ÞCapacity originalð Þ � 100%

ð20Þwhere Capacity(watermarked) and Capacity(original) arethe file capacity of the watermarked image and the ori-ginal image capacity.We can analyze the following conclusions from

Table 1.The CIR of JPEG image decreases when the quality

factor decreases. Even when the quality factor is below90, CIR is lower than 0. When compression factor is re-duced to 90, bit error happens. In tamper detecting,even when the quality factor goes down to 70, no tamperhappens.

3.2 Attacks and analysis of DRM watermark algorithmIn this experiment (see Figs. 10, 11 and 12), we firstly at-tack the watermarked image and then verify whether wecan extract the watermark from the object imageattacked and also whether we can track the tamperedtracks. In the experiment, we gave the detailed experi-ments according to the four kinds of attacks: (1) copyand paste attack, (2) insert a picture attack, (3) insertgraph attack, and (4) insert color inverse attack, wherethe NC of the extracted watermark is 0.9787, 0.9945,0.9891, and 0.9554, respectively.For further experimental verification, we use different

mobile images as experimental objects. The watermarkdescribes the mobile photo information, such as phoneno., time, location, and IMEI, and the attack methodsare the same as above. Table 2 shows the same results asdescribed above and the experiments show the valid-ation and performance of the proposed algorithm.From the results in Fig. 9, the watermark is easily rec-

ognized, which describes the mobile photo information,such as phone no., time, location, and IMEI after theattacks.According to the experimental results from Table 2,

we draw the conclusion as follows: The CIR for dif-ferent images are in general less than 0 when com-pression quality is 90, which means the file size ofwatermarked images cannot be increased very muchfor watermark embedding, even decreased for JPEGcompression. The PSNR of the images are in generalabove 38 dB, and the bit error of the identificationwatermarks are near 0.The related methods are compared with the proposed

schemes in terms of functionalities and performance.Comparing with current related schemes, our pro-

posed scheme has the following three advantages andinnovations.Obviously, in addition to the above-mentioned analysis,

the present schemes have another three advantages.

Fig. 12 Simulation results


1. Automatic watermark generation: Compared withsome other watermark scheme, the watermarkimage is generated automatically, using templatesand photo information.

2. Parameter control for watermark performance: Jpegcompression can control the quality ofwatermarked mobile image, so as to balance thecapacity of the image file and watermarkperformance.

3. Scheme proposed misuse-tracing method based onrobust watermark in post-usage stage: The proposedwatermark scheme supports robust watermark andsemi-fragile watermark, which can certify mobilephone source authentication and detect the integrityand tamper tracking of the mobile pictures, even if thewatermarked image was attacked by variant tamping.

Through experiments, we can see that the proposedwatermark scheme is practical and effective, which cantrack source forensics and recognize the mobile-relatedinformation to trace and find the one who misuse orviolate the confidential mobile pictures. As a highlight ofthe scheme, we can track where the attacks occurred.In Table 3, we can get the watermark algorithm that is

embedded randomly in the host image, which is security.The bit error is 0 even when the tampering ratio is up to2.6%. And the algorithm is effective especially whencolor inverse attack happens; the bit error is very slow inexperiment.

4 ConclusionsWe proposed a new watermark-based mobile image datasecurity scheme to protect confidential image data in thispaper. We embedded some mobile-identification-relatedand user-identification-related information into themobile photo after the mobile photos were taken, such asphone no., time, location, and IMEI of the mobile phone,which were robust watermarks for traceability and source

authentication confirmation. Once the image data wasmisused such as spreading the confidential image inthe Internet or for commercial purposes withoutauthorization, we can identify and trace the misusedresponsibility by extracting the watermark embeddedin the image data. Finally, we evaluated the proposedwatermark scheme for a mobile by groups of variant sizeimage data for security and efficiency, a large amount ofgroups of experiments manifest the proposed scheme wassecure, efficient, pervasive, and robust for confident imagedata protection and misuse tracing.

AbbreviationsDCT: Discrete cosine transfer; DRM: Digital rights management;DWT: Discrete wavelet transfer

AcknowledgementsNot applicable.

FundingThis work is supported by the National Natural Science Fundamental ofChina (No. 61272519, No. 61170297, No. 61572080, No. 61472258).

Availability of data and materialsThe datasets used is based on benchmark dataset Lenna, and the real data isbased on the mobile Internet Phone with the type GIONEE M6.

Authors’ contributionsThe work is mainly finished by author MZ, who designed the watermarkmodel and algorithm and wrote the whole paper. Author JM co-operatedfor the experiments and checking the references. Both authors read andapproved the final manuscript.

Competing interestsThe authors declare that they have no competing interests.

Publisher’s NoteSpringer Nature remains neutral with regard to jurisdictional claims inpublished maps and institutional affiliations.

Author details1Information Security Center of Beijing University of Posts andTelecommunications, Beijing 100876, China. 2Audio and Video department,The Third Research Institute of China Electronics Technology GroupCorporation, Beijing 100015, China.

Received: 19 May 2018 Accepted: 16 July 2018

References1. A Piva, An overview on image forensics. ISRN Signal Processing 2013, 1–22 (2013)2. MC Stamm, M Wu, KJR Liu, Information forensics: an overview of the first

decade. IEEE Access 1, 167–200 (2013)3. L Ozparlak, I Avcibas, Differentiating between images using wavelet-based

transforms: a comparative study. IEEE Trans on Information Forensics andSecurity 6(4), 1418–1431 (2011)

4. H Farid, MJ Bravo, Perceptual discrimination of computer generated andphotographic faces. Digit. Investig. 8(3/4), 226–235 (2012)

5. TY Chang, SC Tai, GS Lin, A passive multi-purpose scheme based onperiodicity analysis of CFA artifacts for image forensics. J. Vis. Commun.Image Represent. 25(6), 1289–1298 (2014)

6. P Fei, L Juan, L Min, Identification of natural images and computergenerated graphics based on hybrid features. International Journal of DigitalCrime and Forensics 4(1), 1–16 (2012)

7. X Wang, L Yong, B Xu, et al., A statistical feature based approach todistinguish PRCG from photographs. Comput. Vis. Image Underst. 128(11),84–93 (2014)

Table 2 Experiments for different images when quality is 90

Image number Image CIR (%) PSNR (dB) Bit error (%)

1 Stadium − 29.57 38.69 0.0001

2 Motto − 29 39.92 0

3 Research Building − 34.98 35.95 0.0001

4 Science park − 29.67 38.98 0.0003

Table 3 Experiments for tampering

Image number Image Tampering ratio (%) Bit error (%)

1 Stadium 5.3 0.0039

2 Motto 1 0

3 Research building 1.4 0.0041

4 Science park 19 0.0015


8. P Fei, L Jiaoting, L Min, Identification of natural images and computer-generated graphics based on statistical and textural features. J. Forensic Sci.60(2), 435–443 (2015)

9. S Gao, Z Cong, C Wu, et al., A hybrid feature based method for distinguishingcomputer graphics and photo-graphic image[G] //LNCS 8398:Proc of the IntWorkshop on Digital-Forensics and Watermarking 2013 (Springer, Berlin,2013), pp. 303–313

10. L Zhaohong, Z Zhenzhen, YQ Shi, Distinguishing computer graphics fromphotographic images using a multiresolution approach based on local binarypatterns. Security and Communication Networks 7(11), 2153–2159 (2014)

11. ALS Orozco, JR Corripio, LJG Villalba, et al., Image source acquisitionidentification of mobile devices based on the use of features. MultimediaTools and Application 75(12), 7087–7111 (2016)

12. C Chen, MC Stamm, Camera model identification framework using anensemble of demosaicing features // Proc of the Int Workshop on InformationForensics and Security (IEEE, Piscataway, 2015), pp. 1–6

13. Z Deng, A Gijsenij, Z Jingyuan, Source camera idenfication using auto-whitebalance approximation // Proc of the Int Conf ono Computer Vision (IEEE,Piscataway, 2011), pp. 57–64

14. K Goyal, R Panwar, N Khanna, Evaluation of IQM’s Effectiveness for Cell PhoneIdentification Using Captured Videos and Images //Proc of the Int Conf onPower, Control and Embedded System (IEEE, Piscataway, 2014), pp. 1–6

15. F Razzazi, Seyedadabi, A robust feature for single image camera identificationusing local binary patterns // Proc of the IEEE Int Symp on Signal Processingand Information Technology (IEEE, Piscataway, 2014), pp. 462–467

16. TH Thai, R Cogranne, F Retraint, Camera model identification based on theheteroscedastic noise model. IEEE Trans. Image Process. 23(1), 250–263 (2014)

17. L Shuhan, X Kong, B Wang, et al., Silhouette coefficient based approach oncell-phone classification for unknown source images //Proc of the 2012 IEEE IntConf on Communications (IEEE, Piscataway, 2012), pp. 6744–6747

18. J Fridrich, Sensor Defects in Digital Image Forensic (Springer, Berlin, 2013),pp. 179–218

19. AR Soobhany, KP Lam, P Fletcher, et al., Mobile camera source identificationwith SVD // LNEE 313: Proc of the Int Joint Conf on Computer Information andSystems Sciences and Engineering (Springer, Berlin, 2015), pp. 123–131

20. C Mo, J Fridrich, M Goljan, et al., Source digital camcorder identification usingsensor photo response non-uniformity //Proc of the 9th Conf on Security,Steganography, and Watermarking of Multimedia Contents IX (SPIE, SanFrancisco, 2007) 65051G–65051G-12

21. G Hongmei, A Swaminathan, M Wu, Intrinsic sensor noise features forforensic analysis on scanners and scanned images. IEEE Trans onInformation Forensics and Security 4(3), 476–491 (2009)

22. B Mahdian, S Saic, Blind authentication using periodic properties ofinterpolation. IEEE Transactions of Information Forensics and Security 3(3),529–538 (2008)

23. AC Popescu, H Farid, Exposing digital forgeries by detecting traces ofresampling. IEEE Transactioni on Signal Processing 53(2), 758–767 (2005)

24. Z Zhen, K Jiquan, P Xijian, Blind detection of image splicing based onimage quality metrics and moment features. Journal of ComputerApplications. 28(12), 3108–3111 (2008)

25. C Chen, YQ Shi, S Wei, in Proc.of International Conference on PatternRecognition. A machine learning based scheme for double JPEGcompression detection (United states: Institute of Electrical and ElectronicsEngineers, Tampa, 2008), pp. 1–11

26. W Luo, Z Qu, J Huang, in Proc.of IEEE International Conference on Acoustics,Speech and Signal Processing. A novel method for detecting cropped andrecompressed image block (Institute of Electrical and Electronics Engineers,Honolulu, 2007), pp. 217–220

27. H Farid, Exposing digital forgeries from JPEG ghosts. IEEE Transactions onInformation Forensics and Security 4(1), 154–160 (2009)

28. S Bayram, I Avcibas, B Sankur, Image manipulation detection. Journal ofElectronic Imaging 15(4), 12–18 (2006)

29. J Fridrich, J Lukas, M Goljan, in Proc.of the International Society for OpticalEngineering. Detecting digital image forgeries using sensor pattern noise(SPIE, San Jose, 2009), pp. 118–126

30. A Popescu, H Farid, Exposing digital forgeries in color filter arrayinterpolated images[J]. IEEE Transactions on Signal Processing. 53(10),3948–3959 (2005)

31. F Di Martino, S Sessa, Fragile watermarking tamper detection with imagescompressed by fuzzy transform. Information Sciences 195(13), 62–90 (2012)

32. S Rawat, B Raman, A chaotic system based fragile watermarking scheme forimage tamper detection. AEU-International Journal of Electronics andCommunications. 62(10), 840–847 (2011)

33. RO Preda, Semi-fragile watermarking for image authentication with sensitivetamper localization in the wavelet domain. Measurement 46(1), 367–373 (2013)

34. Z Lin, J He, X Tang, C-K Tang, Fast, automatic and fine-grained tamperedJPEG image detection via DCT coefficient analysis. Pattern Recognition42(11), 2492–2501 (2009)

35. H-C Wu, C-C Chang, Detection and restoration of tampered JPEG compressedimages. Journal of Systems and Software 64(2), 151–161 (2002)


secure and efficient drm watermark algorithm of forensics

Documents