Computers and Electrical Engineering 40 (2014) 51–58

Contents lists available at ScienceDirect

Computers and Electrical Engineering

journal homepage: www.elsevier .com/ locate/compeleceng

Robust and blind watermarking scheme for three dimensionalanaglyph images q

0045-7906/$ - see front matter � 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.compeleceng.2013.11.005

q Reviews processed and approved for publication by Editor-in-Chief Dr. Manu Malek.⇑ Corresponding author. Mobile: +91 9043453723.

E-mail addresses: ivysaji@gmail.com (I. Prathap), anitha_nadarajan@mail.psgtech.ac.in (R. Anitha).1 Tel.: +91 422 2572177.

Ivy Prathap ⇑, R. Anitha 1

Department of Applied Mathematics & Computational Sciences, PSG College of Technology, Coimbatore 641 004, Tamil Nadu, India

a r t i c l e i n f o a b s t r a c t

Article history:Available online 2 December 2013

In this paper, a robust and blind watermarking scheme for three dimensional (3-D)anaglyph images is proposed. Since the computational cost of Red–Green–Blue (RGB)processing is quite high, we have used 3-D Discrete Wavelet Transform (3-D DWT) todecompose the image and process it directly. The watermark to be embedded is computedfrom the image and also the process involves watermark strength factor which scales thequality of the watermark. Jacket matrix is used due to its simplicity in the watermarkembedding and extraction processes. Experimental results show that the proposed schemeis highly imperceptible and robust against various image processing and signal processingattacks. Comparisons with the state-of-the-art demonstrate the effectiveness of ourscheme.

� 2013 Elsevier Ltd. All rights reserved.

1. Introduction

With the increasing use of 3-D graphics in video games, films, Computer Aided Design (CAD), virtual reality applications,medical imaging, scientific simulation, and cultural heritage, digital entertainment interest has recently moved towards ana-glyph 3-D images. Anaglyph 3-D is the name given to the stereoscopic 3-D effect achieved by means of encoding each eye’simage using filters of different (usually chromatically opposite) colors, typically red and cyan [1]. Anaglyph 3-D images con-tain two differently filtered color images, one for each eye. When viewed through the ‘‘color coded’’ ‘‘anaglyph glasses’’, eachof the two images reaches one eye, revealing an integrated stereoscopic image. The visual cortex of the brain fuses this intothe perception of a three dimensional scene or composition. Anaglyph images have seen a recent resurgence due to thepresentation of images and video on the Internet, Blu-ray Discs, Compact Discs, and even in print. Low cost paper framesor plastic-framed glasses hold accurate color filters that typically make use of all the three primary colors. The current normis red and cyan, with red being used for the left channel. There is a material improvement of full color images, with the cyanfilter, especially for accurate skin tones. Examples from NASA include Mars Rover imaging, and the solar investigation calledSTEREO, which uses two orbital vehicles to obtain the 3-D images of the sun. Other applications include geological illustra-tions by the United States Geological Survey, and various online museum objects. A recent application is stereo imaging ofthe heart using 3-D ultra-sound with plastic red/cyan glasses. Anaglyph images are much easier to view than either parallel(diverging) or crossed-view pair stereograms [1].

Unfortunately, like digital images and audio/video clips, 3-D anaglyph images can be easily duplicated and redistributedwithout any loss of quality by a pirate. This illegal behavior infringes the copyright of graphic model owners and it can alsodo harm to the whole underlying commercial chain. Therefore, a critical demand there now exists on the intellectual

52 I. Prathap, R. Anitha / Computers and Electrical Engineering 40 (2014) 51–58

property protection of these images. Digital watermarking [2] is considered as an efficient solution to solve this emergingproblem. The main idea of watermarking is to hide some secret information which is called the watermark, within the multi-media content such that it is difficult to remove. According to the application, watermarking techniques can be classified intorobust and fragile schemes. Robust watermarking is usually designed for intellectual property protection while fragile water-marking is used for digital content authentication and verification. The design goal of a robust watermarking scheme is tomake the embedded watermark resistant to various attacks. Robustness of the watermarking techniques against severaltypes of attacks is a major challenge in multimedia watermarking. A watermarking scheme in which original image is notneeded for watermark extraction and verification is called a blind scheme. In many of the real time applications, blindschemes are preferable. Depending on the work domain in which the watermark is embedded, watermarking schemescan be branched into two categories: spatial-domain watermarking and transform domain watermarking. In spatial domainwatermarking schemes, the watermark is embedded by directly modifying the pixel values of the image [3,4]. These schemesare less complex but they show poor robustness against common image processing attacks. In transform domain schemes,the watermark is embedded by exploiting the transform properties mainly for watermark imperceptibility and robustness[5,6]. Some of the commonly used transforms for embedding watermarks are Discrete Cosine Transform (DCT), DiscreteWavelet Transform (DWT) and Discrete Fourier Transform (DFT). Wavelet-based approaches have become increasingly pop-ular due to their excellent spatial localization and frequency distribution [7]. The researchers have recently focused on 3-DDWT especially in the field of video and spectral images watermarking [8–11]. These methods are robust to various attacksand are perceptually invisible.

Lot of watermarking schemes have been developed for 3-D models [12,13], Non-uniform Rational Basis Spline (NURBS)surfaces [14], meshes [15,16], DIBR 3-D image representations [17] and stereo images [18]. On the other hand, very littleattention has been given to the development of 3-D anaglyph image watermarking algorithms. A Robust security frameworkfor 3-D images is proposed in [19]. In this paper, the 3-D anaglyph image is converted into RGB channels and each channel isprocessed separately. The watermark embedding is done by means of two transforms, viz. Fractional Fourier Transform andReversible Integer Transform along with Singular Value Decomposition which makes it more complex.

The attenuated aim of this work is to develop a simple but robust watermarking scheme to protect 3-D anaglyph imagesagainst possible image processing and signal processing attacks. The 3-D anaglyph image which is to be watermarked (coverimage) is decomposed using 3-D DWT. Next, the middle level sub bands are divided into blocks and each block is maskedusing Jacket matrix [20,21]. Jacket matrices are a class of matrices with their inverse being determined element-wise. Sinceinverse of Jacket matrix can be calculated easily, it is very helpful in the areas of signal processing, encoding, mobile com-munication, image compression, cryptography, watermarking, etc. [22,23]. The watermark is embedded by changing thediagonal elements of the masked blocks. The experimental results demonstrate better visual imperceptibility, resiliencyand robustness of the proposed scheme against intentional or unintentional variety of attacks in comparison with someof the existing schemes.

The rest of this paper is organized as follows: Section 2 provides the mathematical preliminaries and Section 3 details thewatermark embedding and extraction procedures. Section 4 presents the experimental results which illustrate the imper-ceptibility and robustness of the scheme. Section 5 compares the performance of the proposed scheme with that of somestate-of-the-art schemes. Finally, Section 6 concludes the paper.

2. Mathematical preliminaries

In this section, we present the basic mathematical concepts which are used in the proposed watermarking scheme.

2.1. Multi-level 3-D wavelet transform

The generalized form of 3-D wavelet transform is defined below [24].Let u, v, w be the co-ordinates of the 3-D image and k1; k2; k3 be the corresponding magnitudes. Then

zðu; v;wÞ ¼Xk1�1

i1¼0

Xk2�1

i2¼0

Xk3�1

i3¼0

w1ði1Þ �w2ði2Þ �w3ði3Þ � xð2u� i1;2v � i2;2w� i3Þ ð1Þ

where w1ði1Þ; w2ði2Þ; w3ði3Þ correspond to the filter coefficients and xð2u� i1;2v � i2;2w� i3Þ relates to the scalingparameters.

One of the major challenges in digital watermarking is to achieve a better trade-off between robustness and impercep-tibility. Robustness can be increased by increasing the strength of the embedded watermark, but visual distortion would beincreased as well. In this work, DWT is much preferred because it provides both a simultaneous spatial localization and afrequency spread of the watermark within the cover image. The proposed method applies multilevel 3-D Discrete WaveletTransform by using Haar wavelet transform. The main motivation of using multilevel 3-D Wavelet Transform is that there isno requirement to divide the original image into RGB planes and process separately. The transform efficiently handles the3-D anaglyph image directly.

I. Prathap, R. Anitha / Computers and Electrical Engineering 40 (2014) 51–58 53

2.2. Jacket matrix

A Jacket matrix is defined as below [22]:If the inverse of a square matrix J ¼ jkl½ �n�n is obtained element-wise, then it is said to be a Jacket matrix.

J ¼

j11 j12 . . . j1n

j21 j22 . . . j2n

..

. ... . .

. ...

jn1 jn2 . . . jnn

0BBBB@

1CCCCA

ð2Þ

J�1 ¼ 1n

1j11

1j12

. . . 1j1n

1j21

1j22

. . . 1j2n

..

. ... . .

. ...

1jn1

1jn2

. . . 1jnn

0BBBBBB@

1CCCCCCA

T

ð3Þ

The main benefit of using Jacket matrix in the proposed scheme is masking the original middle level 3-D DWT sub bandcoefficients before embedding the watermark and removal of the mask in the extraction phase efficiently due to the simplic-ity of the Jacket matrix.

3. Proposed watermarking scheme

The proposed algorithm uses 3-D DWT and Jacket matrix. First the cover image is transformed using multi level 3-D DWTand the middle level sub bands are divided into blocks. In the next step, the Jacket matrix is applied to the middle level subband blocks and the watermark is embedded by modifying the diagonal elements of each block. The level of decompositionof 3-D DWT, block size, minimum value of middle level sub bands and watermark strength factor are used in the embeddingprocess which increases the security of the watermark. Finally, inverse 3-D DWT is performed to construct the watermarked3-D anaglyph image.

3.1. Watermark embedding

Transform domain methods lead to stronger watermark embedding while keeping the quality of the watermarked imageat an acceptable level. Insertion of watermarks should not introduce visible distortion in the cover image. Having this inmind, we embed the watermark into the blocks of middle level sub bands since they are less sensitive to watermark manip-ulation. The objective of this phase is to embed a secret watermark reliably in the cover anaglyph image. The functional blockdiagram of the embedding procedure is shown in Fig. 1.

3.1.1. Watermark generationThe watermark to be embedded is computed from the image and it incorporates the watermark strength factor. Since the

quality of the watermark is scaled by the watermark strength factor, it is ideal to use it in the watermark generation.

Fig. 1. Watermark embedding process.

54 I. Prathap, R. Anitha / Computers and Electrical Engineering 40 (2014) 51–58

Steps:

1. Choose a positive integer l; 1 6 l 6 6:2. Apply l-level 3-D DWT to the cover image.3. Divide the middle level sub bands into n� n blocks.4. Find the minimum value of middle level sub bands s.5. Choose the watermark strength factor a:6. Compute the watermark as.

3.1.2. Embedding procedureMultiply each block with the Jacket matrix for masking the original middle level sub bands of l-level 3-D DWT.

D ¼ JBJ�1 ð4Þ

where B denotes the block and J denotes a Jacket matrix of order n� n and J�1 denotes the inverse of J.Then, linear embedding of the watermark as is done by modifying the diagonal elements of each block as below:

Eði; iÞ ¼ Dði; iÞ þ as ð5Þ

where Dði; iÞ; 1 6 i 6 n are the diagonal elements of each block. Eði; iÞ denotes the watermark embedded block. Finally, in-verse 3-D DWT is performed to generate the watermarked image.

3.2. Watermark Extraction

The objective of this phase is to extract the watermark from the watermarked image. The watermark extraction is donewith the help of the secret key ðl;n; s;aÞ. Even if an attacker is aware of the embedding procedure, he cannot correctly extractthe watermark and restore the original image since the watermark extraction depends on multiple secret factors. This in-creases the security of the scheme. Fig. 2 depicts the watermark extraction process.

Steps:

1. Perform l-level 3-D wavelet transform to the watermarked image.2. Partition the middle level sub bands into non-overlapping n� n blocks.3. Compute

Dði; iÞ ¼ Eði; iÞ � as ð6Þ

where Dði; iÞ are the diagonal elements of the extracted block and Eði; iÞ denotes the watermark embedded block.4. Multiply each block with Jacket matrix J and the inverse of the Jacket matrix J�1.

B ¼ JDJ�1 ð7Þ

so that the original diagonal elements of the block are restored back.5. Perform inverse l-level 3-D DWT to recover the watermark and the image.

Only the intended receiver can retrieve the watermark correctly provided that he knows ðl;n; s;aÞ. All the values havesame impact on the extraction since a single modified value leads to faulty extraction.

Fig. 2. Watermark extraction process.

I. Prathap, R. Anitha / Computers and Electrical Engineering 40 (2014) 51–58 55

4. Experimental results and discussions

The experiments were carried out in a laptop equipped with Intel Core (2) Duo processor with 3 GB memory and a clockspeed of 2.10 GHz. In order to explore the performance of the proposed watermarking algorithm, MATLAB platform is usedand a number of experiments were performed on different original 3-D anaglyph images such as Tsukuba, Cone, Dolls, Baby,Teddy, Venus and Flower of size 400� 256 that are obtained using linear projection algorithm considering red–cyan colors[25]. For the sake of comparison, the proposed method is tested on 3-D mesh models such as Venus, Horse, Bunny andDragon also. In our experiments, 6-level 3-D Haar wavelets are used and the resultant middle level sub bands of the 3-Danaglyph images are divided into 2� 2 sub blocks. Watermark strength factor a is chosen as 0.0097981 and a Jacket matrixof order 2� 2 is generated. The performance of the proposed watermarking method is investigated in terms of impercepti-bility, robustness and security. Higher PSNR and correlation coefficient values show better imperceptibility and robustness.

4.1. Imperceptibility

The imperceptibility between the original image and its watermarked counterparts show that the embedded watermarkis visually transparent and watermark embedding does not result in perceptible distortion. The proposed watermarkingalgorithm is run on the test images. Figs. 3a and 3b show some original and watermarked 3-D anaglyph images respectively.In our experiments, for all test images, we could obtain PSNR values greater than 51 dB which reflect high imperceptibility ofthe method and the values are shown in Table 1.

4.2. Robustness

Robustness represents the resistance to various types of intentional and unintentional image processing and signalprocessing attacks that may lead to the distortion or removal of watermark. Correlation coefficient is used as a metric toquantify resistivity of a watermarking scheme against various attacks. The correlation coefficient q between the embeddedwatermark w and extracted watermark �w is defined as:

qðw; �wÞ ¼P

iðwðiÞ �wmeanÞð �wðiÞ � �wmeanÞffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPi wðiÞ �wmeanð Þ2

Pi

�wðiÞ � �wmeanð Þ2q ð8Þ

If two watermarks are identical, then its value will be +1, if they are completely opposite, then its value will be �1 and itwill be 0, if watermarks are completely uncorrelated. Watermarked images are attacked by using JPEG compression, low-pass filtering, median filtering, Gaussian noise, scaling, cropping and rotation respectively. After subjecting to the attacks,watermarks are extracted and correlation coefficients are computed. Table 2 summarizes the experimental results forproposed watermarking scheme against various attacks. From the table, one can understand that the method is highly robustsince all correlation coefficients are greater than 0.9.

We have tested the performance of the proposed technique against several signal processing attacks like Gaussian noiseaddition, filtering and JPEG compression. Addition of noise is a method to estimate the robustness of the watermark.Generally, addition of noise is responsible for the degradation and distortion of the image. The watermarked images aretested for assessing the robustness to Gaussian noise with a variance of 0.5. The results shown in Table 2 demonstrate thatproposed algorithm is robust to the additive noise. The digital image watermarking should be robust to image processing

(a) Tsukuba (b) Cone (c) Dolls

(d) Teddy (e) Venus (f) Flowerpots

Fig. 3a. Test 3-D anaglyph images.

(g) Tsukuba (h) Cone (i) Dolls

(j) Teddy (k) Venus (l) Flowerpots

Fig. 3b. Watermarked 3-D anaglyph images.

Table 1PSNR values of 3-D anaglyph watermarked images.

Images Tsukuba Cone Dolls Teddy Venus Baby1 Flowerpots

PSNR (dB) 53.98 51.83 51.05 56 53 52.5 55

Table 2Correlation coefficients of different 3-D anaglyph images against attacks.

Attacks Tsukuba Cone Dolls Baby1 Flowerpots

Watermark (without any attack) 1 1 1 1 1Median filter (2 � 2) 0.9898 0.9910 0.9937 0.9411 0.9712Averaging filter (7 � 7) 0.9401 0.9517 0.9514 0.9716 0.9516Sharpening (3 � 3) 0.9989 0.9979 0. 9999 0.9963 0.9981Color quantization 0.9537 0.9157 0.9449 0.9510 0.9505Gaussian filtering 0.9980 0.9432 0.9670 0.9657 0.9511Gaussian noise (scale = 0.50) 0.9583 0.9420 0.9499 0.9599 0.9799JPEG compression (CR = 10,50,100) 1 0.9999 1 0.9989 0.9994Median filter (2 � 2) + JPEG 90 0.9920 0.9799 0.9819 0.9818 0.9727Median filter (3 � 3) + JPEG 90 0.9896 0.9989 0.9779 0.9826 0.9210Sharpening (3 � 3) + JPEG 90 0.9473 0.9999 0.9945 0.9321 0.9010Gaussian filter (3 � 3) + JPEG 90 1 0.9999 0.9967 0.9912 0.9989Rotation (50%) 0.9989 0.9887 0.9812 0.9915 0.9975Scaling 0.9689 0.9700 0.9796 0.9656 0.9334Translation 1 1 1 1 1Resizing 0.9689 0.9559 0.9876 0.9543 0.9503Cropping 0.9799 0.9629 0.9754 0.9541 0.9334Contrast adjustment 0.9800 0.9718 0.9760 0.9818 0.9799Histogram equalization 0.9919 0.9809 0.9912 0.9880 0.9918

56 I. Prathap, R. Anitha / Computers and Electrical Engineering 40 (2014) 51–58

techniques such as filtering. Three filters are tested against watermarked images viz. Gaussian filter, median filter with 2 � 2window and averaging filter. Experimental results in Table 2 show that the proposed algorithm is robust against filtering.Another most common manipulation in digital image is image compression. To check the robustness against image compres-sion, the watermarked image is tested with JPEG compression attacks of compression ratios 10, 50 and 100. The correlationcoefficient of extracted watermarks is very close to 1, which emphasizes its robustness. We also tested our proposed water-marking method for histogram equalization, rotation and scaling. From Table 2, it is clear that the proposed algorithm resistperfectly to histogram equalization since the correlation coefficient values are greater than 0.98. Also the algorithm resistsrotation attack. We have extracted watermark from 50� rotated watermarked image and the computed correlation coeffi-cient values for the images Tsukuba, Dolls, Cone, Baby and Flowerpots are 0.9989, 0.9887, 0.9812, 0.9915 and 0.9975respectively.

To fit the image into the desired size, enlargement or reduction is commonly performed and resulted in information lossof the image including embedded watermark. For this attack, the size of the watermarked image is reduced to half and again

Table 3Embedding/extraction times of the proposed method.

Images Embedding time (s) Extraction time (s)

Tsukuba 1.8222 0.1886Cones 2.0876 0.2316Teddy 2.7066 0.3216Venus 1.7750 0.1652

Table 4Comparison of correlation coefficients of extracted watermarks after attacks.

Images Proposed method Bhatnagar et al. [19] Liu and Tan [2]

Cone Tsukuba Dolls Cone Tsukuba Dolls Cone Tsukuba Dolls

No attack 1 1 1 1 1 0.9999 1 0.9998 1Average filtering (7 � 7) 0.9514 0.9401 0.9517 0.8503 0.8543 0.8422 0.7560 0.7502 0.7547Median filtering (7 � 7) 0.9937 0.9898 0.9910 0.9406 0.9357 0.9372 0.8091 0.8171 0.8003Gaussian noise (50%) 0.9499 0.9583 0.9420 0.8491 0.8514 0.8435 0.8169 0.8118 0.8157JPEG compression (CR = 50) 1 1 0.9999 0.9719 0.9763 0.9757 0.9440 0.9497 0.9475Cropping (50%) 0.9754 0.9799 0.9629 0.9649 0.9586 0.9558 0.6472 0.6420 0.6441Resizing 0.9876 0.9689 0.9559 0.9475 0.9455 0.9424 0.7511 0.7678 0.7517Rotation (50�) 0.9812 0.9989 0.9887 0.8909 0.8970 0.8891 0.6838 0.6887 0.6853Histogram equalization 0.9912 0.9919 0.9809 0.9597 0.9548 0.9593 0.9593 0.9672 0.9508Contrast adjustment 0.9760 0.9800 0.9718 0.9768 0.9777 0.9744 0.9517 0.9599 0.9537Sharpening (50%) 0.9999 0.9989 0.9979 0.9956 0.9963 0.9964 0.9906 0.9876 0.9931

Table 5Comparison of correlation coefficients after Gaussian noise with mesh watermarking.

Ampitude (%) Proposed method Wang et al. [16]

0.10 0.30 0.50 N.unif.30 N.unif.50 0.10 0.30 0.50 N.unif.0.30 N.unif.0.50

Venus 0.98 0.97 0.87 0.91 0.88 0.94 0.87 0.78 0.89 0.73Horse 0.98 0.95 0.89 0.95 0.90 0.98 0.86 0.77 0.92 0.78Bunny 0.98 0.98 0.89 0.95 0.91 0.98 0.85 0.77 0.95 0.85Dragon 0.97 0.96 0.88 0.85 0.89 0.98 0.76 0.61 0.72 0.53

I. Prathap, R. Anitha / Computers and Electrical Engineering 40 (2014) 51–58 57

brought to its original size. The correlation coefficient values for all watermarked images are exactly 1 and shown in Table 2.The combination of different image processing and signal processing attacks are also done to test the robustness of the algo-rithm. We have tested the watermarked images with median filter combined with JPEG compression, sharpening and Gauss-ian noise combined with JPEG compression. The results are quite inspiring and shown in Table 2. The embedding andextraction procedures are very fast and Table 3 illustrates the embedding and extraction times for four test images.

5. Comparison with existing schemes

In order to demonstrate the significant performance of the proposed scheme, a more elaborated performance comparisonwith the existing methods [2,16,19] is done. The watermarking technique in Gaurav [19] is the only 3-D anaglyph imagewatermarking scheme available. They have implemented the method given by Liu and Tan [2] for 3-D anaglyph imagesand compared with their scheme.

Our comparative analysis uses the same cover images of size 400 � 256. We have compared our proposed scheme withthe watermarking schemes of 3-D anaglyph images by Gaurav [19] and Liu and Tan [2] and Table 4 gives the results. Table 4clearly shows that the proposed method performs better than the existing schemes. We have also compared the perfor-mance of our method with a robust 3-D mesh watermarking method by Wang et al. [16]. Table 5 compares the robustnessagainst random noise addition between 3-D mesh watermarking method [16] and the proposed method. The abbreviation‘‘N.unif.’’ in Table 5 specifies the non-uniform noise. Our scheme performs better than the said method against random noise.Moreover, the method [16] is not robust to cropping attack. Simulation results indicate that our method outperforms theexisting robust anaglyph image watermarking methods in a wide range of attacks.

6. Conclusion

In this paper a new and simple watermarking technique for 3-D anaglyph images has been presented. The proposed algo-rithm takes the advantage of multilevel 3-D DWT instead of complex RGB processing techniques as in the traditional meth-

58 I. Prathap, R. Anitha / Computers and Electrical Engineering 40 (2014) 51–58

ods. Also in this method, the level of 3-D DWT decomposition, block size, minimum value of middle level sub bands andwatermark strength factor are used as secret keys. So, even if an attacker knows the embedding and extraction procedures,without knowing the secret keys, one cannot identify the location where the watermark is embedded and therefore cannotextract the watermark correctly. Moreover this is a blind watermarking scheme which does not need the original image forextraction. The proposed method achieves good imperceptibility with PSNR values higher than 51 dB. Our experimental re-sults show that the proposed scheme is robust against JPEG compression, Gaussian noise, median filtering, sharpening, rota-tion, cropping, resizing, histogram equalization and many other image processing and signal processing attacks. Also, it iscompared with the watermarking scheme in [19] which is the only scheme for 3-D anaglyph images and our schemeachieves better image quality and robustness. Our proposed method is highly robust compared to 3-D mesh watermarkingscheme [16]. The experimental results clearly demonstrate the improved performance in terms of imperceptibility androbustness against various attacks. Also the proposed method is very fast and hence it could be applied efficiently to protect3D anaglyph images.

Acknowledgements

This work is supported by Tata Consultancy Services, (TCS), Chennai, India. The authors are thankful to TCS for providingfinancial support for doctoral research work. The authors also thank the editor and the anonymous reviewers for their valu-able comments for improving the quality of this paper.

References

[1] 3D anaglyph images; November 2013. <http://en.wikipedia.org/wiki/Anaglyph_3D>.[2] Liu R, Tan T. An SVD-based watermarking scheme for protecting rightful ownership. IEEE Trans Multimedia 2002;4(1):121–8.[3] Nikolaidis N, Pitas I. Robust image watermarking in the spatial domain. Signal Process 1998;66(3):385–403.[4] Yu P-T, Tsai H-H, Lin J-S. Digital watermarking based on neural networks for color images. Signal Process 2001;81(3):663–71.[5] Lin SD, Shie S-C, Guo J. Improving the robustness of DCT-based image watermarking against JPEG compression. Comput Stand Inter 2010;32(1):54–60.[6] Peng H, Wang J, Wang W. Image watermarking method in multiwavelet domain based on support vector machines. J Syst Softw 2010;83(8):1470–7.[7] Al-Haj A. Combined DWT–DCT digital image watermarking. J Comput Sci 2007;3(9):740.[8] Galiano V, López-Granado O, Malumbres M, Migallón H. Fast 3D wavelet transform on multicore and many-core computing platforms. J Supercomput

2013:1–18.[9] Kim S-J, Kim T-S, Kwon K-R, Ahn S-H, Lee K-I. Digital watermarking based on three-dimensional wavelet transform for video data. In: Advances in

multimedia information processing-PCM 2005. Springer; 2005. p. 337–48.[10] Campisi P, Neri A. Perceptual video watermarking in the 3D-DWT domain using a multiplicative approach. In: Digital watermarking. Springer; 2005. p.

432–43.[11] Kaarna A, Krasavin K, Parkkinen J, Jaaskelainen T, van Schyndel R. Watermarking spectral images with three-dimensional wavelet transform subject to

various illumination conditions. J Imag Sci Technol 2008;52(3). 30502–1.[12] Wang W-B, Zheng G-Q, Yong J-H, Gu H-J. A numerically stable fragile watermarking scheme for authenticating 3D models. Comput Aided Des

2008;40(5):634–45.[13] Chou C-M, Tseng D-C. A public fragile watermarking scheme for 3D model authentication. Comput Aided Des 2006;38(11):1154–65.[14] Pan J, Zheng J, Zhao G. Blind watermarking of NURBS curves and surfaces. Comput Aided Des 2013;45(2):144–53.[15] Choi H-I, Kim T-w, Kwon S-H, Moon HP, Park SH, Shin H-J, et al. Digital watermarking of polygonal meshes with linear operators of scale functions.

Comput Aided Des 2010;42(3):163–72.[16] Wang K, Lavoué G, Denis F, Baskurt A. Robust and blind mesh watermarking based on volume moments. Comput Graph 2011;35(1):1–19.[17] Lin Y-H, Wu J-L. A digital blind watermarking for depth-image-based rendering 3D images. IEEE Trans Broadcast 2011;57(2):602–11.[18] Yu M, Wu A, Luo T, Jiang G, Zhou W, Fu S. A new stereo image watermarking method for 3D media. Procedia Eng 2012;29:2399–404.[19] Bhatnagar G, Wu J, Raman B. A robust security framework for 3D images. J Visual 2011;14(1):85–93.[20] Lee MH, Hou J. Fast block inverse jacket transform. IEEE Signal Process Lett 2006;13(8):461–4.[21] Lee MH, Finlayson K. A simple element inverse Jacket transform coding. IEEE Signal Process Lett 2007;14(5):325–8.[22] Lee MH. Jacket matrix and its applications to signal processing. In: 2011 IEEE 10th international conference on trust, security and privacy in computing

and communications (TrustCom). IEEE; 2011. p. 5–6.[23] Lee MH, Zhang X-D, Song W, Xia X-G. Fast reciprocal Jacket transform with many parameters. IEEE Trans Circ Syst I: Regular Pap 2012;59(7):1472–81.[24] Mohanty BK, Meher PK. Parallel and pipeline architectures for high-throughput computation of multilevel 3-D DWT. IEEE Trans Circ Syst Video

Technol 2010;20(9):1200–9.[25] Test images; November 2013. <http://vision.middlebury.edu/stereo/data>.

Ivy Prathap received the B.Sc. Computer Science degree in 1998 and M.Sc. Computer Science degree in 2004 from Mahatma Gandhi University, Kottayam,Kerala. Currently, she is working towards the Ph.D.degree in Computer Applications at Department of Applied Mathematics and Computational Sciences,PSG College of Technology, Coimbatore. Her research area is Digital Watermarking.

R. Anitha is an Associate Professor with the Department of Applied Mathematics and Computational Sciences, PSG College of Technology, Coimbatore, India.She received her M.Sc. degree in Mathematics from Madurai Kamaraj University, Madurai and got her M.Phil. and Ph.D. degrees in Mathematics fromBharathiar University, Coimbatore. Her research interests include Cryptography, Information Security, Digital Watermarking, Botnet and Malware Detec-tion.