a robust and optimized 3d red-cyan anaglyph blind image · pdf file ·...

Contemporary Engineering Sciences, Vol. 9, 2016, no. 32, 1575 - 1589HIKARI Ltd, www.m-hikari.com

https://doi.org/10.12988/ces.2016.69156

A Robust and Optimized 3D Red-Cyan Anaglyph

Blind Image Watermarking in the DWT Domain

Hidangmayum Saxena Devi

National Institute of Technology ManipurTakyelpat Imphal

Manipur, India

Khumanthem Manglem Singh

National Institute of Technology ManipurTakyelpat Imphal

Manipur, India

Copyright © 2016 Hidangmayum Saxena Devi and Khumanthem Manglem Singh. This

article is distributed under the Creative Commons Attribution License, which permits un-

restricted use, distribution, and reproduction in any medium, provided the original work is

properly cited.

Abstract

Robust anaglyph image watermarking provides a way for copyrightprotection of 3D red-cyan anaglyph image by hiding a secret message ora logo into an anaglyph image. In this work, a novel robust optimizedblind encrypted three dimensional red-cyan anaglyph image watermark-ing system is developed in which the original red-cyan anaglyph coverimage is decomposed using DWT and a binary image of size 32 × 32is embedded after applying AES encryption method into the three di-mensional anaglyph image utilizing BPN to train the optimized coeffi-cients obtained after application of DWT. DWT is utilized because of itsgood space-frequency localization and lower computational cost. Theproposed scheme comprises of two modules: training phase and testingphase. In the training phase, the 3D anaglyph image is formed by linearprojection method from the left stereo and right stereo images. In thetraining phase, discrete wavelet transform is applied to the 3D anaglyphimage to get the low pass and high pass filter value and optimization ofthe DWT transformed bits (LH and HL) is done using genetic algorithm

1576 Hidangmayum Saxena Devi and Khumanthem Manglem Singh

(GA) and are trained using back propagation neural network. On theother hand, during the testing phase, the binary image of size 32×32 isencrypted using Advanced Encryption Standard method and is embed-ded by adding it to the optimized transformed bits and extraction ofthe embedded image is done finding the Eigen feature vectors that givesthe characteristic values for the extraction of the watermarked image.Based on the experiment conducted using Middlebury stereo dataset,it can be concluded that the proposed work gives satisfactory resultscompared to the existing 3D anaglyph image watermarking schemes tilldate.

Keywords:DWT (Discrete Wavelet Transform), BPN (Back PropagationNetwork), Manifold Harmonics Transform (MHT), Depth Image Based Ren-dering (DIBR), 3-D Discrete Wavelet Transform (3D-DWT), GA (Genetic Al-gorithm)

1 Introduction

Due to the advancement in the Internet and various techniques in 3D tech-nology like 3D digitization, 3D printing etc., illegal use of 3D images, videosetc. which include duplication, redistribution, and unauthorized access areincreasing rapidly. So, these activities continuously give a threat to the own-ers causing a great loss [16]. In order to take action and precaution againstthese offenses, copyright protection or the content protection or authentica-tion of such 3D images or videos has become a serious and indispensable task.And, one of the feasible solution to this is the application of 3D watermark-ing which uses the technique of hiding information either vividly or secretlyinto 3D models like triangular meshes, polygonal models, meshes, 3D imagesetc. The hidden message can be embedded either in the spatial domain or inthe frequency domain with the difference that the message is embedded usingthe pixel bits directly in the spatial domain and on the other hand, in thefrequency domain, the message is embedded in the transformed coefficients ofthe image. Frequency domain techniques are more robust than spatial domaintechniques. Robust watermarking system should be able to withstand againstimage processing attacks and is used for copyright protection etc. The robust-ness of a watermarking system is determined by the normalized correlation(NC) coefficient between the watermarked image and the original image. Thehigher the NC value, the better the robustness of the system.

Many researchers have already worked on 3D watermarking schemes whichcan be grouped according to the type of 3D data used which include 3D models[11,12,15,23,29,33], 3D mesh models [6,7,13,14,26,27], polygonal mesh models[8, 10, 20,22], 3D images [21,24,25,28,30,32] etc.

A robust and optimized 3D red-cyan anaglyph blind image watermarking 1577

AES encryption based schemes have been used with other transformationtechniques have been proposed by some of the researchers [1, 5]. In [1], AESencryption sheme have been applied to the watermark formed by the con-catenation of the compressed R-S vector, hash value prepared using encryptedMD5 is prepared from the input DICOM image and patient ID. In [5], a water-marking algorithm alongwith encryption was proposed in which the encryptionis done using AES and watermarking is performed using LSB and QIM.

Anaglyph image watermarking has been proposed by researchers [4, 25].In [4], Bhatnagar et.al has implemented anaglyph image watermarking usingFractional Fourier Transform and Reversible Integer Transform. He has furthercompared his scheme with Liu and Tan’s method [9]. Ivy et.al [25] have alsoimplemented the watermarking of anaglyph images and produced more robustsystem than Bhatnagar’s method [4].

The main aim of this paper is to produce a more robust 3D red-cyananaglyph image watermarking scheme. In our proposed scheme, the origi-nal 3D anaglyph cover image is formed from its left and right stereo pair takenfrom Middlebury stereo data set [17] and is transformed using DWT and theDWT transformed bits are optimized using genetic algorithm which are fedas inputs to BPN to train the optimized bits. The binary image is encryptedusing AES algorithm and is embedded using the optimized bits. The blindwatermark extraction method is done finding the matched feature values forthe watermarked image. It is observed from the experimental results that theproposed system is robust against attacks viz. rotation, translation, cropping,impulse noise, Gaussian noise, histogram equalization, JPEG compression etc.

The paper is organized as follows: Section 2 discusses the related theoriesand mathematical concepts regarding genetic algorithm, BPN, AES. Section3 describes the proposed watermarking system and experimental results andcomparison with the other existing systems are discussed in Section 4 andSection 5 respectively and finally conclusion is discussed in Section 6.

2 Theoretical Background

In this section, related theories and applied methods are discussed.

2.1 Genetic Algorithm Overview

Genetic Algorithm are adaptive methods used in solving search and optimiza-tion problems. The basics of GA was discussed by Holland [3]. GA's operateon a number of feasible solutions under the application of fitness function pro-ducing better and optimized solutions. The basic GA consists of the followingoperations:


i. Initialization of population: population is the number of individuals. Pop-ulation can be randomly chosen.

ii. Evaluation of the fitness of each chromosome in the population.

iii. Creation of new population repeating the steps viz. selection, crossover,mutation and accepting the new population.

iv. The new formed population is used for further run of the GA algorithm.

v. The algorithm is stopped when the end condition is satisfied retuning thebest solution in the current population.

In our proposed scheme, GA is used to optimize the lower and upper boundbits of a DWT transformed 3D anaglyph image in order to obtain the signifi-cant bits for the embedding process by using the fitness function.

Fitness Function = @ fitness fn (e, Fs, Ft), is defined in Equation 1.

fitness fn =

{1 ifFs < Ft

0 otherwise(1)

where, e deals with error probability, Fs is the current feature and Ft, thetotal feature vector.

The optimized bits are fed as input to BPN for training and to select thedesired locations for embedding.

2.2 Back Propagation Network (BPN)

Back propagation neural network is a systematic method of training multi-layer artificial neural networks. The main objective of BPN is for optimizingthe weights so as to enable the neural network learn how to correctly maparbitrary inputs to outputs. It has the advantage in reconstructing missingdata when the original information is incomplete and also finds the set ofweights that minimizes the error.

A BPN consists of at three layer of units viz. input layer, intermediatehidden layer and output layer. The input units are connected to units in thehidden layer; the hidden layer units are connected fully to output layer unitsin a feed-forward manner.

The proposed back propagation network architecture consists of 1-10-1which means that it consists of 1 input neuron, 10 hidden neurons and 1 out-put neuron, in which the input layer consists of the optimized bits obtainedfrom genetic algorithm described above as its input using tan-sigmoid hiddenlayer for reducing the errors producing the desired output.


2.3 Advanced Encryption Standard Overview

Due to the misuse of data by several unauthenticated persons, there arisessituations to protect data from such unauthenticated users. Encryption isone of the methods to provide security and confidentiality and for protectingprivacy. Data protection is possible by means of data encryption enabling usersto protect their data by rendering it incomprehensible to any person who isnot intended for the purpose. AES [2] is one of the cryptographic methodswhich is as shown in Fig.1. It consists of the following main steps [18]:

i Initialization of AES components: In this step, the S-box and the polyno-mial matrix are created.

ii Substitute bytes:In this particular step, replacement of each byte by thebyte indexed by row (left 4 bits) and column (right 4 bits) of a 16 × 16table.

iii Shift Rows: During this step, the last three rows of the state matrix iscyclically shifted keeping the first row unchanged.

iv Mix Columns: The columns of the state matrix is transformed via a four-term polynomial.

v Add round key: The current round key (matrix) is added to the state(matrix) using XOR operation.

3 Proposed watermarking methodology

The proposed watermarking system involves two phases: training phase andtesting phase. The three dimensional anaglyph image is formed from theircorresponding left stereo pair and the right stereo pair image of Middleburystereo dataset by linear projection technique. The proposed training phaseand testing phase are discussed below.

3.1 Training phase framework

The proposed training phase is described below:

i. The 3D anaglyph image is transformed using l DWT decomposing intofour bands LL, LH, HL and HH.

ii. The transformed bits of subbands LH and HL are optimized using geneticalgorithm.

iii. Lastly, the optimized bits are trained using back propagation neural net-work.


Input Watermark Image

Convert each bit into binary

Initialize AES Components

XOR first Round Key

Initialize AES Components

Sub Bytes

Shift Rows

Mix Columns

XOR nth Round Key

Sub Bytes

Shift Rows

XOR nth Round Key

Encrypted Watermark

Last Round

Steps are repeatedfornine times

Figure 1: AES algorithm

3.2 Testing phase framework

The testing phase comprises of the following.

i. It consists of GA optimization on the upper and lower bits of DWT trans-formed image.

ii. The binary watermark is encrypted using AES algorithm as given in Fig.1and embedded using the flowchart given in Fig.2.

iii. Extract the watermark using the flowchart given in Fig.3


Load Cover image

Find the sizeof the

cover image

Generate matrix for rgbcomponents of the coverimage to generate zerosmatrix of the equal sizes

for the embedding process

Find the sizeof the

embedded image

Issize of

embedded image)<

size ofcover image

Embedding and Extraction is not done

Start the iteration toadd the bits of the

cover image locationsas covered by sizeof the cover image

Add the arrays of theboth the cover image

and encryptedwatermark image

Watermarked 3Danaglyph image

No

Yes

Figure 2: Proposed embedding phase

Load theWatermarked

image

Find the feature pointsof the watermarked imageand also of the watermarkimage using the minimum

eigen value algorithm

Extract those locationswith

matched feature points

Extractthe

watermark

Figure 3: Proposed extraction phase

4 Results and Discussion

4.1 Database used

The proposed 3D Anaglyph watermarking system is tested using the Middle-bury stereo dataset [17]. The 3D red-cyan anaglyph image is formed from thecorresponding stereo pair of image using the linear projection technique. A setof some test red-cyan anaglyph images used in our experiment are shown inFig. 4 alongwith their respective left and right stereo pair of images from thedataset. The system works on the original cover image of size 370 × 460 × 3.Some of the watermarked anaglyph images are shown in Fig.5.

4.2 Performance Analysis

The performance of our proposed system is evaluated in terms of impercep-tibility and robustness against various intentional image processing attacks


which are shown in Table I and Table II respectively. The imperceptible na-ture of our proposed system is illustrated in Table I. It is observed from theobtained experimental results that the system is highly perceptible obtainingPSNR values above 51 dB, which signifies that there will be no visible distor-tion between the original image and the watermarked image. The proposedsystem is also evaluated for its robust nature against various intentional im-age processing attacks viz. median filtering, average filtering, Gaussian noise,scaling, translation, rotation, blurring, color quantization, gamma correction,histogram equalization and cropping. The robustness of any watermarkingsystem is determined from the NC value, determined by Equation 2.

ρ(y, Y ) =

∑i[y(i) − ymean][Y (i) − Y mean]∑

i[y(i) − ymean]2[Y (i) − Y mean]2(2)

where y and Y are the original and the extracted watermark images. Ifthey are identical, ρ = 1, different if ρ = −1 and completely uncorrelated ifρ = 0.

(a) (b) (c) (d) (e)

(a) (b) (c) (d) (e)

Figure 4: .(a) Left Stereo image (b) Right stereo Image (c) Generation of 3DAnaglyph image (d) original watermark (e) Recovered/Extracted Watermark.


(a) (b) (c)

Figure 5: Sample of watermarked 3D Anaglyph images.

Table 1: PSNR values of some of the watermarked 3D Anaglyph Images.Sl.No. Images PSNR Sl.No. Images PSNR Sl.No. Images PSNR

1 Doll 53.4634 7 Aloe 55.2739 13 Monopoly53.27162 Cones 53.398 8 Wood1 53.4165 14 Cloth1 53.01513 Tsukuba51.5628 9 Flowerpot 53.2124 15 Rocks1 53.09154 teddy 53.398 10 Lampshade171.8377 16 Bowling1 58.0275 Venus 53.3324 11 Midd1 53.48856 Baby1 55.5176 12 Plastic 53.071

1584Hidan

gmayum

Sax

enaDevian

dKhuman

them

Man

glemSingh

Table 2: Normalized correlation coefficient of the different 3D Anaglyph images after attacks.Attacks Aloe Baby1Flowerpot1Lampshade1Midd1 Dolls RocksWood1PlasticMonopoly

Averaging Filtering([7 × 7])0.999830.99991 0.99994 0.99997 0.999890.999850.99984 0.99989 0.99984 0.99985Median Filtering([7 × 7]) 0.999830.99991 0.99994 0.99999 0.999890.999860.99984 0.99988 0.99983 0.99986

Sharpening(50%) 0.999910.99994 0.99994 1 0.999910.999910.99989 0.99991 0.99984 0.99992Color Quantization 0.999840.99994 0.99993 0.99997 0.999890.999860.99986 0.99992 0.99979 0.99986

Gaussian Noise 0.999910.99987 0.99993 0.99999 0.99991 0.9999 0.99989 0.99996 0.99986 0.99995JPEG Compression 0.999820.99989 0.99993 1 0.999880.999870.99989 0.99987 0.99982 0.99985

Rotation(50°) 0.999780.99981 0.99982 0.99986 0.999790.999690.99978 0.99986 0.9999 0.99989Cropping(50%) 0.999990.99988 0.99994 1 0.999980.999960.99996 0.99987 0.99991 0.9999

Blurring 0.9998 0.99988 0.99993 1 0.999870.999860.99984 0.99986 0.99982 0.99983Scaling 0.999760.99982 0.99995 0.99992 0.999890.99995 0.9999 0.9998 0.99974 0.99985

Translation 0.999820.99986 0.9998 0.9999 0.999790.999820.99995 0.99991 0.99986 0.99994Resizing 0.9998 0.99983 0.99996 0.99994 0.999950.999960.99991 0.99986 0.99989 0.99987

Impulse Noise 0.999990.99999 0.99999 1 0.999990.999990.99998 0.99999 0.99997 0.99999Histogram Equalization 0.99981 0.9999 0.99995 0.99997 0.999880.999860.99987 0.99987 0.99989 0.99984

No Attack 1 1 1 1 1 1 1 1 1 1

Arob

ust

andop

timized

3Dred

-cyanan

aglyphblin

dim

agewaterm

arking

1585

Table 3: Comparison of proposed watermarking scheme with the existing 3D red-cyan anaglyph Image watermarkingschemes.

Proposed method Bhatnagar Liu and Tan IVYImages Cones Tsukuba Dolls ConesTsukubaDolls ConesTsukubaDolls ConesTsukubaDolls

No Attack 1 1 1 1 1 1 1 1 1 1 1 1Averaging filter([7 × 7]) 0.99999 0.9998 0.99985 0.8503 0.8543 0.8422 0.7560 0.7502 0.7547 0.9514 0.9401 0.9517Median Filtering([7 × 7]) 1 0.99981 0.99986 0.9406 0.9357 0.9372 0.8091 0.8171 0.8003 0.9937 0.9898 0.9910

Gaussian Noise 0.99998 0.99988 0.9999 0.8491 0.8514 0.8435 0.8169 0.8118 0.8157 0.9499 0.9583 0.9420JPEG Compression 0.99998 0.99988 0.9999 0.9719 0.9763 0.9757 0.9440 0/9497 0.9475 1 1 0.9999

Cropping 0.99997 0.99972 0.99996 0.9649 0.9856 0.9558 0.6472 0.6420 0.6441 0.9754 0.9799 0.9629Resizing 0.99999 0.99994 0.99996 0.9475 0.9455 0.9424 0.7511 0.7678 0.7517 0.9876 0.9689 0.9559

Rotation(50◦) 0.99977 0.99999 0.99969 0.8909 0.8970 0.8891 0.6838 0.6887 0.6853 0.9812 0.9989 0.9887HE 0.99999 0.99985 0.99986 0.9597 0.9548 0.9593 0.9593 0.9672 0.9508 0.9912 0.9919 0.9809

Sharpening 0.99999 0.99986 0.99991 0.9956 0.9963 0.9964 0.9906 0.9876 0.9931 0.9999 0.9989 0.9979


4.3 Comparison with the existing systems

The proposed red-cyan anaglyph watermarking scheme is compared for robust-ness with the existing related systems [4, 9, 25]. It is evident from the resultslisted in Table 3 that the system is robust against various intentional attackslike Gaussian noise, translation, scaling, rotation, gamma correction, medianfiltering, average filtering, resizing, cropping, blurring etc.

When the system is tested with Gaussian noise of scale 0.5, the systemachieves correlation coefficient value more than 0.9999 which reveals its ro-bustness nature.

The proposed system is also tested against various attacks such as rotationwith 50◦, it achieves NC value above 0.99969 which is almost closed to 1indicating its capacity to withstand against rotation.

When the system is also tested with scaling attack, it shows its robustnessachieving the NC value above 0.99974.

The system can withstand various other attacks such as rotation, scaling,translation, blurring, impulse noise, cropping, rotation, JPEG compression,histogram equalization, color quantization, median filtering, average filteringetc., it achieves NC values of 0.9999, closely related to 1. Hence, the proposedred-cyan anaglyph watermarking scheme is more robust as compared to theother existing ones.

5 Conclusion

In this paper, a novel, robust, imperceptible and efficient 3D Anaglyph imagewatermarking method is implemented which utilizes optimized DWT trans-form and BPN. The locations for the feature points of the eigen features areused for blind extraction. The proposed method achieves PSNR values above51dB and robust against various attacks viz.median filtering, average filter-ing, Gaussian noise, cropping, sharpening, histogram equalization etc. Theproposed method is compared with the various existing 3D Anaglyph water-marking schemes and the experimental analysis clearly shows that it performsbetter than these existing schemes.

References

[1] Mohamed M. Abd-Eldayem, A proposed security technique based onwatermarking and encryption for digital imaging and communicationsin medicine, Egyptian Informatics Journal, 14 (2013), no. 1, 1-13.https://doi.org/10.1016/j.eij.2012.11.002


[2] Ross Anderson, Eli Biham and Lars Knudsen, Serpent: A proposal for theadvanced encryption standard, NIST AES Proposal, 174 (1998), 1-23.

[3] David Beasley, R. R. Martin and D. R. Bull, An overview of geneticalgorithms: Part 1. Fundamentals, University Computing, 15 (1993), 58-69.

[4] Gaurav Bhatnagar, Jonathan Wu and Balasubramanian Raman, A robustsecurity framework for 3D images, Journal of Visualization, 14 (2011), no.1, 85-93. https://doi.org/10.1007/s12650-010-0067-5

[5] Dalel Bouslimi, Gouenou Coatrieux and Christian Roux, A joint encryp-tion/watermarking algorithm for verifying the reliability of medical im-ages: Application to echographic images, Computer Methods and Pro-grams in Biomedicine, 106 (2012), no. 1, 47-54.https://doi.org/10.1016/j.cmpb.2011.09.015

[6] X. Q. Feng and Yanan Liu, A robust, blind and imperceptible watermark-ing of 3D mesh models base on redundancy information, Int. J. DigitalContent Technol. Appl., 6 (2012), no. 2, 172-179.https://doi.org/10.4156/jdcta.vol6.issue2.21

[7] Chen-Tsung Kuo, et al., A blind robust watermarking scheme for 3Dtriangular mesh models using 3d edge vertex detection, Asian Journal ofHealth and Information Sciences, 4 (2009), no. 1, 36-63.

[8] Chao-Hung Lin, et al., A novel semi-blind-and-semi-reversible robust wa-termarking scheme for 3D polygonal models, The Visual Computer, 26(2010), no. 6, 1101-1111. https://doi.org/10.1007/s00371-010-0461-y

[9] Ruizhen Liu and Tieniu Tan, An SVD-based watermarking schemefor protecting rightful ownership, IEEE Transactions on Multimedia, 4(2002), no. 1, 121-128. https://doi.org/10.1109/6046.985560

[10] R. Ohbuchi, A. Mukaiyama and S. Takahashi, A frequency-domain ap-proach to watermarking 3D shapes, Computer Graphics Forum, 21(2002), no. 3, 373-382. https://doi.org/10.1111/1467-8659.t01-1-00597

[11] Singh, Law Kumar, Deepak Chaudhry, and Gopalji Varshneya, A NovelApproach of 3D Object Watermarking Algorithm using Vertex Normal,International Journal of Computer Applications, 60 (2012), no. 5.

[12] Xinyu Wang, Yongzhao Zhan and Shun Du, A Non-blind Robust Water-marking Scheme for 3D Models in Spatial Domain, Chapter in Electri-cal Engineering and Control, Springer Berlin Heidelberg, 2011, 621-628.https://doi.org/10.1007/978-3-642-21765-4 76


[13] Stefanos Zafeiriou, Anastasios Tefas and Ioannis Pitas, A blind ro-bust watermarking scheme for copyright protection of 3D mesh mod-els, IEEE International Conference on Image Processing, 3 (2004).https://doi.org/10.1109/icip.2004.1421366

[14] Yang Liu, Balakrishnan Prabhakaran and Xiaohu Guo, Blind invisiblewatermarking for 3D meshes with textures, 17th IEEE International Con-ference on Image Processing (ICIP), (2010).https://doi.org/10.1109/icip.2010.5652524

[15] Sharvari C. Tamane and Ratnadeep R. Deshmukh, Blind 3D Model Wa-termarking Based on Multi-Resolution Representation and Fuzzy Logic,4 (2012), 117-126. https://doi.org/10.5121/ijcsit.2012.4110

[16] https://www.jisc.ac.uk/guides/3d-digitisation-and-intellectual-property-rights

[17] http://vision.middlebury.edu/stereo/data

[18] http://www.facweb.iitkgp.ernet.in/ sourav/AES.pdf

[19] S. Cai and X. Shen, Octree-based robust watermarking for 3D model,Journal of Multimedia, 6 (2011), no. 1, 83-90.

[20] Ying Yang and Ioannis Ivrissimtzis, Polygonal mesh watermarking usingLaplacian coordinates, Computer Graphics Forum, 29 (2010), 5.https://doi.org/10.1111/j.1467-8659.2010.01767.x

[21] Abbas, Tawfiq A., Majid Jabbar Jawad and Sud Sudirman, Robust Wa-termarking of Digital Vector Maps for Copyright Protection.

[22] Y. Fu, Robust image watermarking scheme based on 3D-DCT, Sixth In-ternational Conference on Fuzzy Systems and Knowledge Discovery, 5(2009), 437-441. https://doi.org/10.1109/fskd.2009.19

[23] Rakhi C. Motwani and Frederick C. Harris Jr., Robust 3D WatermarkingUsing Vertex Smoothness Measure, IPCV, (2009), 287-293.

[24] Hee-Dong Kim, Ji-Won Lee, Tae-Woo Oh, Heung-Kyu Lee, Robust DT-CWT watermarking for DIBR 3D images, IEEE Transactions on Broad-casting, 58 (2012), no. 4, 533-543.https://doi.org/10.1109/tbc.2012.2206851

[25] Ivy Prathap and R. Anitha, Robust and blind watermarking scheme forthree dimensional anaglyph images, Computers & Electrical Engineering,40 (2014), no. 1, 51-58.https://doi.org/10.1016/j.compeleceng.2013.11.005


[26] Emil Praun, Hugues Hoppe and Adam Finkelstein, Robust mesh water-marking, Proceedings of the 26th Annual Conference on Computer Graph-ics and Interactive Techniques, (1999).https://doi.org/10.1145/311535.311540

[27] Han Sae Song, Nam Ik Cho and Jong Weon Kim, Robust watermarkingof 3D mesh models, IEEE Workshop on Multimedia Signal Processing,(2002). https://doi.org/10.1109/mmsp.2002.1203313

[28] Ying Wang, Xue-feng Zheng and Hai-Yan Liu, Robust 3D Watermark-ing Based on Geometry Image, 4th IEEE International Conference onWireless Communications, Networking and Mobile Computing, (2008).https://doi.org/10.1109/wicom.2008.778

[29] Mitrea, Mihai, Toward robust spread spectrum watermarking of 3D data,(2004).

[30] A. Tefas, G. Louizis and I. Pitas, 3D image watermarking robust to ge-ometric distortions,IEEE International Conference on Acoustics, Speechand Signal Processing (ICASSP), 4 (2002), IV-3465.

[31] Thomas Harte and Adrian G. Bors, Watermarking 3D models, Interna-tional Conference on Image Processing, 3 (2002).https://doi.org/10.1109/icip.2002.1039057

[32] S. J. Jaipuria, Watermarking for Depth Map Based 3D images usingwavelet transform, International Conference on Communications and Sig-nal Processing, (2014), 181-185.https://doi.org/10.1109/iccsp.2014.6949824

[33] A. Kalivas, A. Tefas and I. Pitas, Watermarking of 3D models using prin-cipal component analysis, International Conference on Multimedia andExpo, 1 (2012), 637-640.

[34] Min-Su Kim, Sebastien Valette, Ho-Youl Jung, Remy Prost, Watermark-ing of 3D irregular meshes based on wavelet multiresolution analysis,Chapter in Digital Watermarking, Springer Berlin Heidelberg, 2005, 313-324. https://doi.org/10.1007/11551492 24

Received: October 2, 2016; Published: December 3, 2016

a robust and optimized 3d red-cyan anaglyph blind image · pdf file ·...

Documents