Nâng cao hiệu quả và hiệu năng giấu tin trong ảnh số TT tieng anh

MINISTRY OF EDUCATION AND TRAINING CAN THO UNIVERSITY VÕ PHƯỚC HƯNG ENHANCING THE CAPACITY AND EFFICIENCY OF DATA HIDING IN DIGITAL IMAGES SUMMARY OF DOCTORAL THESIS MAJOR: INFORMATION SYSTEMS Major Code: 62480104 CAN THO - 2020 The thesis is completed at: CAN THO UNIVERSITY Advisor: Assoc Dr Do Thanh Nghi The thesis will be defended on the meeting of Univeristy doctoral thesis evaluation council Place: Doctoral thesis evaluation hall, 2nd floor – Administration Building, Can Tho University At hour day month year 2020 The dissertation is available at: Learning Resource Center, Can Tho University and The National Library of Vietnam PUBLICATIONS OF DOCTORAL THESIS [CT1] Nguyễn Thái Sơn, Võ Phước Hưng, Huỳnh Văn Thanh, Đỗ Thanh Nghị, “Giấu tin thuận nghịch ảnh Stereo với khả nhúng tin cao”, Kỉ yếu hội thảo FAIR 2016, pp 631-637 [CT2] P-H Vo, T-S Nguyen, V-T Huynh and T-N Do A robust hybrid watermarking scheme based on DCT and SVD for copyright protection of Stereo images in proc of NAFOSTED Conf on Information and Computer Science (NICS 2017), 2017, pp 331-335 [CT3] P-H Vo, T-S Nguyen, V-T Huynh and T-N Do A Novel Reversible Data Hiding Scheme with Two-Dimensional Histogram Shifting Mechanism in International Journal of Multimedia Tools and Applications, Vol.77(21): 2877728797, Springer, 2018 [SCIE] [CT4] Võ Thành C, Võ Phước Hưng, Trầm Hoàng Nam, Nguyễn Thái Sơn, Đỗ Thanh Nghị, "Một thuật toán thủy vân ảnh số mạnh dựa DWT, DCT, SVD đặc trưng SIFT", Kỉ yếu hội thảo FAIR 2019 [CT5] P-H Vo, T-S Nguyen, V-T Huynh, T-C Vo and T-N Do Secure and Robust Watermarking Scheme in Frequency Domain Using Chaotic Logistic Map Encoding in proceeding of International Conference on Computer Science, Applied Mathematics and Applications (ICCSAMA 2019), Springer, 2019 [Indexed by Scopus] [CT6] P-H Vo, T-S Nguyen, V-T Huynh, T-N Do A High capacity invertible steganography method for Stereo image in the book Digital Media Steganography: Principles, Algorithms, Advances, ELSEVIER Inc., 2020 [Scopus] CHAPTER 1: INTRODUCTION 1.1 Motivation The digital transformation evolution has brought about profound changes in our society and our lives More and more digital information is communicated over the network A lot of important information is being exchanged which has attracted the attention of many unauthorized users such as snooping, masquerading or modification Therefore, protecting the security of such information is of considerable concern to many researchers For decades, the researchers have tried to propose methods to avoid the attention of malicious attackers Therefore, the security of the transmitted digital information is guaranteed Unlike cryptography, data hiding techniques are the science of embedding information into cover digital multimedia, such as images, video, and audio The cover image after being embedded with secret data, called stegoimage, is still in a meaningful format Therefore, data hiding techniques can avoid attracting attackers’ attention 1.2 Data hiding mechanisms in digital images Along with powerful software, new device, such as digital cameras and camcorder, high-quality scanner, have reached consumers worldwide to create, manipulate and enjoy multimedia data Thus, in this research, digital images use as cover objects Moreover, the content of the image also needs to protect the copyright and authentication of the entire The block diagram presentation of a data hiding system is depicted in Fig 1.1 Fig 1.1 Block diagram of a data hiding system where: • • • • • • • IC: Cover Image S: Secret message k: Key fEmbed: Embedding function IS: Stego image fExtract: Extracting function I′C: Retrieval image Embedding procedure, secret message S is embedded into IC based on fEmbed and k Which is formulated as follow 𝐼𝑆 = 𝑓𝐸𝑚𝑏𝑒𝑑 (𝐼𝐶 , 𝑆, [𝑘]) (1.1) Extracting procedure, the secret message S is retrieved from stego-image IS based on fExtract and k It can be defined as the following equation (1.2) (𝑆, 𝐼𝐶′ ) = 𝑓𝑒𝑥𝑡𝑟𝑎𝑐𝑡 (𝐼𝑆 , [𝑘]) (1.2) The embedding process has been well done such that the stego-image IS looks like the cover image IC The difference between IS and IC is called distortion Embedded information is extracted from IS image which is based on embedding function Embedding capacity and human visual quality efficiency are evaluated on the properties of the data hiding system Belong to the individual application, the priority of the properties is a different requirement The properties of data hiding may be: • • • • • Imperceptibility Hiding capacity Fidelity Robustness Security 1.3 Research scopes The capacity and efficiency of data hiding in digital images belong to the embedding algorithm The system is designed such as a high embedding capacity is embedded into the cover image as well as stego-image is maintaining in terms of human visual quality Moreover, the system is robust against attacks First, the effectiveness of the embedding scheme can be evaluated by answering questions: (1) Can secret messages embedded in the image be safe? (2) (2) Does the stego-image retain the same quality as the original image? (3) Is it sure that hidden information is not discovered? Embedding rate is the second requirement for a data hiding system to increase the number of embedded messages However, embedding capacity and efficiency of the data hiding in a digital image is always the opposite In other words, the more information is embedded, the worse the stego-image quality and vice versa Therefore, it is necessary to study and figure out methods that offer reversibility and high capacity while maintaining an acceptable quality of Stereo images 1.4 Research Objectives The problem identified in this thesis is to transmit confidential information on unsafe transmission channels and prevent illegal copies of copyrighted content for the Stereo image In cases, the cryptography technique is not suitable, data hiding is a good alternative method to protect data The main goal of the thesis is to study and propose efficient schemes of data hiding in Stereo images To archive the goal, the thesis focuses on the following specific objectives: • Overview of solutions for hiding information in images, analyzing, summarizing the strengths and weaknesses of each solution • Research on Stereo digital images, compressed images in the spatial domain, and frequency domain to improve analysis, synthesize how to embed information into images • Research and improve the efficiency and visual stego-image quality as well as retrieve the original image • Research and propose the robust and secure data hiding schemes for copyright protection of Stereo image 1.5 Research methodology The thesis uses analysis, evaluation, synthesis, and experimental demonstration methods and Mathematical theory, image processing, coding, statistics, and optimization are applied to solve research objectives The proposed methods are proven and implemented on Matlab software 1.6 Contribution of the thesis The thesis proposes several new approaches to improve the efficiency and performance of hiding information in Stereo images The main contribution of the thesis is improving and developing techniques for hiding information which are invertible system with high embedding capacity in the frequency domain of images In addition, the schemes are secure and robust with resistance to image processing attacks The thesis has contributed new methods in data hiding, which properly archive the trade-off between embedding rate and human visual quality and security and robustness • A novel reversible data hiding scheme with two-dimensional histogram shifting mechanism, • A High capacity invertible steganography algorithm using 2-D histogram shifting with EDH • A robust hybrid watermarking scheme based on DCT and SVD for copyright protection of Stereo images, • A secure and robust watermarking scheme in the frequency domain using chaotic logistic encoding 1.7 Thesis structure The thesis is organized into five chapters and an appendix as the following Chapter 1: Introduction of information secure and confidential information area How to distinguish the cryptography technique and data hiding technique Chapter 2: General Overview of the state of data hiding, this chapter discusses the idea behind information hiding with respect to reversible and nonreversible steganography and watermarking Chapter 3: This Chapter proposes a reversible data hiding technique based on two-dimensional histogram shifting for quantized discrete cosine transformation coefficients (QDCT) The scheme, a two-dimensional histogram is constructed by QDCT coefficients blocks with the size of 8×8 of the left image and the right image The QDCT coefficients are selected for embedding data to achieve high embedding capacity Chapter 4: A robust hybrid watermarking scheme, secure and robust watermarking schemes are performance which are embedded a logo into image for copyright protection of Stereo image Chapter 5: Conclusion and future works CHAPTER 2: GENERAL OVERVIEW OF STATE OF DATA HIDING 2.1 Introduction The growth of digital information, especially the image which is supported image processing industry, has generated the popularity and availability of digital images Thereby, drawing the deep interest of researchers in using images as an object to hide information in communication As well, digital images are easily manipulated, copied and distributed illegally Therefore, in some cases, the image also needs protection In order to address these challenges, more recently, studies on hidden techniques with different approaches have been conducted To get an overview of the state of data hiding in digital images, the next section in this chapter will introduce and analyze the criteria for assessing the hidden systems After that, studies on information-hiding techniques will be reviewed, analyzed and evaluated 2.2 Evaluation metrics for data hiding system 2.2.1 Image visual quality PSNR is a measure of the visual quality of a stego-image it is measured in decibels (dB) and is one of the most popular techniques to see how much the stego-imge resembles the original image The PSNR is given as equation (2.1) 𝑃𝑆𝑁𝑅 = 10 log10 𝐵2 𝑀𝑆𝐸 (2.1) where, B is the maximum color boundary of the image, MSE is a deviation of the stego-image from the original image which is given by the mean square error: 𝑀 𝑀𝑆𝐸 = 𝑁 ∑ ∑(𝐼𝑠 (𝑥, 𝑦) − 𝐼𝑐 (𝑥, 𝑦))2 𝑀×𝑁 (2.2) 𝑥=1 𝑦=1 with M, N are dimensions of the cover (original) image IC and IS represents the stego-image ; IC(x,y), IS(x,y) are pixel value at the coordinate (x,y) of IC and IS, respectively 2.2.2 The robustness of watermarked image To evaluate the robustness property of the watermarking techniques, the bit correlation error (BCR) is used to measure the correction ratio of the extracted watermark image that is defined as equation (2.3) 𝐵𝐶𝑅 = ′ ̅̅̅̅̅̅̅̅̅̅ ∑𝑛×𝑛 𝑖=1 𝑤𝑖 ⨁𝑤𝑖 𝑛×𝑛 (2.3) where 𝑤𝑖 𝑤𝑖′ are the ith binary value or the original watermark and of the extracted watermark, respectively The notation ⨁ indicates an exclusive-OR operator 2.3 Classification of data hiding techniques The essential feature of data hiding is to keep communication secure while transmitting stego-image over the network or communication channel Different methods for the system were proposed based on the application and stages included in the embedding process Fig 2.1 illustrated a flowchart of classification data hiding techniques Fig 2.1 Flowchart of classification data hiding techniques generated by using DCT and JPEG quantization matrix Each block Bm,n [m ∈ (0, M-1), n∈(0, N-1)] can be divided into three categories named (a) searching sector, which contains some lower frequency QDCT coefficients, (b) embedding sector, which contains some middle-frequency QDCT coefficients and (c) nonused sector, which contains the rest of coefficients 3.2.2.2 Two-dimensional histogram shifting based lossless Data embedding The 2-D histogram denoted as h(x, y), where x and y are the frequencies of two feature values, belong to the QDCT left and right block Taking advantage of the similarities of the left and right images in the stereo images, 2D histogram shifting is used for embedding data bits into cover stereo-images, we explored the coefficients of QDCT located in the embedding area (i.e., ≤ u + v ≤ 9) quietly approaching zero or positive value Based on our scheme, the larger the number of zero coefficients is, the greater the embedding capacity is obtained Moreover, the trend of these coefficients commonly gathers at the top-right corner of the histogram (i.e., NE direction) Thus, the 2D shifting and data embedding are conducted at this corner (see Fig 3.1) to achieve high embedding capacity while maintaining the good stego-stereo image quality as shown in the experimental results (a) Before shifting (b) After shifting 11 (c) After embedding Fig 3.1 The 2D histogram used in the proposed scheme 3.2.3 Experimental results In the application of our research, the threshold 𝓉 is used to choose a pair of similar block pairs for data embedding Additionally, in the experiment, we used various quality factors μ to observe the impact on the stego-stereo image quality and embedding capacity Observing four tested images with the threshold 𝓉 = 20 and μ = 75, the capacity of our method is more swiftly decrease than that one when it reaches the high PSNR However, the proposed scheme is always superior to Yang and Chen’s scheme in terms of embedding capacity and image quality (a) PSNR and EC of Img1 image (b) PSNR and EC of Img33 image 12 (c) PSNR and EC of Img36 image (d) PSNR and EC of Img38 image Fig 3.2 Performance comparison between our scheme with two-dimension histogram shifting and Yang et al.’s method at 𝓉 = 20, μ = 75 We compared the proposed scheme’s performance with those of three other existing frequency-based hiding schemes (Parah et al.’ scheme1, Gou et al.’s scheme2 and Yang et al.’s scheme3) Table demonstrated that the embedding rate of Yang and Chen’s scheme is significantly higher than those of the two other schemes (Parah et al and Gou et al.) However, the embedding rate of Yang and Chen’s scheme (0.131bpp) is still lower than that of the ĐX1’scheme (0.160 bpp) The main reason is that Yang and Chen’s scheme has applied 1D histogram shifting to embed approximately 1.6 secret bits into a pair of QDCT coefficients in stereo images when the different value of these coefficients is equal to zero However, most of the QDCT coefficient pairs in the embedding area of the left and the right images have the values of (0, 0) Therefore, the embedding capacity of Yang and Chen’s scheme is limited by the number of pairs (0, 0) In contrast, the 2D histogram shifting is applied in the proposed scheme to further improve the embedding rate By doing so, for each pair Parah, S.A., Sheikh, J.A., Loan, N.A., Bhat, G.M.: Robust and blind watermarking technique in DCT domain using inter-block coefficient differencing Digital Signal Processing 53, 11–24 (2016) Guo, J., Zheng, P., Huang, J.: Secure watermarking scheme against watermark attacks in the encrypted domain Journal of Visual Communication and Image Representation 30, 125–135 (2015) Yang, W.-C., Chen, L.-H.: Reversible DCT-based data hiding in stereo images Multimed Tools Appl 74, 7181–7193 (2015) 13 with the values of (0, 0), one or two secret bits are embedded in the proposed scheme Table Performance comparison between our scheme with some existing schemes with 𝓉 = 5, μ = 75 Schemes Average Average embedding PSNR (dB) rate (bpp) Domain Invertibility Parah et al 41.48 0.015 DCT không Gou et al 39.82 0.015 DWT+DCT khơng Yang et al 42.35 0.131 DCT có ĐX1 42.40 0.160 DCT có 3.3 A High capacity invertible steganography algorithm using 2-D histogram shifting with EDH 3.3.1 Related works 3.3.2 The proposed method In this scheme, we propose a novel invertible steganography scheme with high embedding capacity for the stereo image using two-dimensional histogram shifting in transform domain with embedding direction histogram (EDH) Which is abbreviated as ĐX1 This is motivated by the fact that DCT-quantized coefficients of each pair similar block in the left and right image of stereo image commonly meet to zero value The DCT-quantized coefficient pairs mentioned here are chosen for embedding secret bits By using EDH to shift the middlefrequency DCT-quantized coefficients of the two-dimensional histogram, the proposed method achieves the trade-off between imperceptibility and embedding capacity The experimental results indicated that our method had better performance compared to related works 3.3.2.1 Generation of the embedding direction histogram (EDH) The embedded secret data is first to split into 3-bit groups and represented in decimal form to make a sequence S={s1, s2, …, s|n|/3}, thus si is in a range from to digits Next, a frequency histogram is built by counting the occurrence of si number Then, the histogram is arranged in descending order of frequency si 14 3.3.2.2 Based-2-D histogram shifting with EDH Data embedding 2-D histogram, a kind of statistical analysis, can present the distribution of 𝐿 QDCT coefficients in blocks: left block (𝐵𝑚,𝑛 ) and similarity block (𝑆𝑖𝑚𝐵𝑚,𝑛 ) Embedding data into the middle frequency zone (4 ≤ u+v ≤ 7) causes fewer impacts on the image quality compared to hiding data into the low-frequency zone Histogram shifting: Before data embedding, the 2-D histogram is shifted except for the coefficients with the value at (-1, -1) coordinate Since EDH is generated by only digits from ‘0’ to ‘7’ with peak digit at the origin coordinate This task aims to expand the histogram for making enough vacant space to hide secret data A mathematical model can be expressed as follows: - if ( |𝐵𝑄𝐿 (𝑢, 𝑣)| = |𝐵𝑄𝑠𝑖𝑚 (𝑢, 𝑣)|) ℌ2(𝐵̂𝑄𝐿 (𝑢, 𝑣), 𝐵̂𝑄𝑠𝑖𝑚 (𝑢, 𝑣)) = (𝐵𝑄𝐿 (𝑢, 𝑣), 𝐵𝑄𝑠𝑖𝑚 (𝑢, 𝑣) + 1) 𝑖𝑓 (𝐵𝑄𝐿 (𝑢, 𝑣) ≥ 1)&(𝐵𝑄𝑠𝑖𝑚 (𝑢, 𝑣) ≥ ) {(𝐵𝑄𝐿 (𝑢, 𝑣) − 1, 𝐵𝑄𝑠𝑖𝑚 (𝑢, 𝑣) + 1) 𝑖𝑓 (𝐵𝑄𝐿 (𝑢, 𝑣) ≤ −1)&(𝐵𝑄𝑠𝑖𝑚 (𝑢, 𝑣) ≥ 1) (3.1) (𝐵𝑄𝐿 (𝑢, 𝑣) + 1, 𝐵𝑄𝑠𝑖𝑚 (𝑢, 𝑣) − 1) 𝑖𝑓 (𝐵𝑄𝐿 (𝑢, 𝑣) ≥ 1)&(𝐵𝑄𝑠𝑖𝑚 (𝑢, 𝑣) ≤ −1) - otherwise 𝐿 , 𝐶 𝑅 + 1) 𝑓𝑜𝑟 {𝐶 𝐿 = 𝑎𝑛𝑑 𝐶 𝑅 ≥ } (𝐶𝑢,𝑣 𝑢,𝑣 𝑢,𝑣 𝑢,𝑣 𝐿 , 𝐶̂ 𝑅 ) = ℎ2 (𝐶̂𝑢,𝑣 𝑢,𝑣 𝐿 + 1, 𝐶 𝑅 )𝑓𝑜𝑟 {𝐶 𝐿 ≥ 𝑎𝑛𝑑 𝐶 𝑅 = 0} (𝐶𝑢,𝑣 𝑢,𝑣 𝑢,𝑣 𝑢,𝑣 𝐿 , 𝐶 𝑅 − 1) 𝑓𝑜𝑟 {𝐶 𝐿 = 𝑎𝑛𝑑 𝐶 𝑅 ≤ −1} (𝐶𝑢,𝑣 𝑢,𝑣 𝑢,𝑣 𝑢,𝑣 (3.2) 𝐿 − 1, 𝐶 𝑅 ) 𝑓𝑜𝑟 {𝐶 𝐿 ≤ −1 𝑎𝑛𝑑 𝐶 𝑅 = 0} { (𝐶𝑢,𝑣 𝑢,𝑣 𝑢,𝑣 𝑢,𝑣 Data embedding: With each coefficient pair (𝐵𝑄𝐿 (𝑢, 𝑣), 𝐵𝑄𝑠𝑖𝑚 (𝑢, 𝑣)) equal to (0,0) with (4 ≤ u+v ≤ 7) secret bits are embedded through the EDH and stego pair is determined by { 𝐵̂𝑄𝐿 (𝑢, 𝑣) =𝑎 𝑠𝑖𝑚 𝐵̂𝑄 (𝑢, 𝑣) = 𝑏 (3.3) 3.3.3 Experimental results Fig 3.3 shows the embedding capacity (EC) performance of 50 stereo pairs with threshold 𝓉 =20 and various JPEG quality factors (μ= 50, 55, 60, 65, 70, 75, 80, 85, 90) The μ parameter affects DCT coefficients after quantization which 15 is a critical feature to reflect the payload of the method which is illustrated in Fig 3.3 Fig 3.3 The payload performance of 50 stereo images is influenced by the μ Furthermore, it is observed from Fig 3.4, the obtained curve corresponding to the μ is better performance with smaller μ value In particular, when μ=50, the EC can achieve high performance Since a lot of QDCT coefficient pairs are reaching to zero in the middle frequency (a) Img1 Image (b) Img2 Image Fig 3.4 EC versus PSNR graph using different values of μ for six stereo pairs Fig 3.5 is shown that the effective embedding capacity of ĐX2’s scheme compared to the embedding capacity of ĐX1’s scheme and Yang et al.’s scheme 16 Fig 3.5 Performance comparison between the ĐX2’s scheme with ĐX1 and Yang et al.’s scheme Table 3.1 shows the comparison of PSNR among ĐX1, ĐX2 and Yang et al.’s method It is obvious that the proposed method achieves the highest PSNR on average among three schemes when the EC is 20,000 bits and the μ is 75 Table 3.1 Comparisons of PSNR in dB between ĐX1, ĐX2 and Yang et al.’s scheme for EC=20,000 bits with μ=75 Image Yang et al ĐX1 ĐX2 Img3 39.86 40.48 40.80 Img13 41.32 42.13 42.76 Img15 41.56 42.45 43.23 Img17 41.31 42.06 43.33 Img21 40.22 40.81 41.37 Average 40.85 41.59 42.30 17 3.3.4 Discussions In this chapter, we proposed two invertible steganography methods for Stereo images which are implemented in the DCT domain The proposed scheme, 2-D histogram based on the QDCT coefficient pairs is plotted to hide secret data bits The proposed solution offers reversibility and high embedding capacity while maintaining an acceptable quality of stereo-images Moreover, to improve the embedding capacity, a high capacity invertible steganography algorithm using the 2-D histogram shifting and the EDH is proposed In the proposed scheme, the EDH is constructed from the secret data to maintain the minimum modification for data embedding Besides, a threshold 𝓉 and μ selection strategy are also proposed to further enhance the embedding performance 18 CHAPTER WATERMARKING IN TRANSFORM DOMAIN FOR COPYRIGHT PROTECTION STEREO IMAGES 4.1 Introduction Digital watermarking is a technique that is widely used to protect rightful ownership of digital images the goal of watermarking is used to protect the copyright and authenticate the entire of watermarked images Watermarking can be classified into fragile watermarking and robust watermarking The fragile watermark is sensitive to any small modification in the watermarked image Thus, it is used for content authentication and tamper detection The robust and secure watermarking schemes maintain the existence of watermark in the watermarked images even when such watermarked images are attacked by image processing operations Therefore, robust and secure watermarking is applied to prevent illegal copies of copyrighted content 4.2 A robust hybrid watermarking scheme based on DCT and SVD for copyright protection of Stereo images 4.2.1 Related works 4.2.2 proposed scheme Watermarking embedding procedure: Watermark embedding processing is classified into three stages In the first stage, the searching for similar blocks is applied as was done in ĐX1’s method In the second stage, a binary watermark is implanted in the SVD transformation of both anti-diagonal DCT similar blocks A matrix A sized of 4×4 is formed by two anti-diagonals, a matched block pair Block A then is transformed into three matrices U, S and V based on SVD transformation Next, the binary watermark image is partitioned into blocks w of size 4×4 and embedded into matrix S as follows: Sw = S + σ × w 19 (4.1) where σ is the robustness factor of the embedded watermark Later, perform an SVD on Sw to get three matrices U1, S1, and V1 such that SVD(Sw) = U1 × S1×V1 (4.2) In the final stage, the inverse transformations are performed to obtain the watermarked stereo image Watermarking extracting procedure: Generally, the watermarked image can be extracted from the watermarked Stereo image The computational steps of the extraction process are implemented by the following steps: - Step 1: perform SVD operation on a matrix 𝐴̌ to decompose it into three ̌, 𝑆̌ and 𝑉̌ such that matrices 𝑈 ̌ 𝑆̌ 𝑉̌ 𝑇 𝐴̌ = 𝑈 (4.3) - Step 2: Calculate the possibly distorted 𝑆′𝑤 𝑆𝑤′ = 𝑈1 𝑆̌ 𝑉1𝑇 (4.4) - Step 3: Extract watermark 𝑤′ 𝑤′ = 𝑆𝑤′ − 𝑆 𝜎 (4.5) 4.2.3 Experimental results Fig 4.1 illustrated the performance of the proposed scheme with σ =10 in terms of the image quality and the BCR under various attacks on the test image “Aloe” As can be seen in the figure even though the watermarked image is attacked by cropping 6.25% or 25%, the BCRs of both restored watermark images are still larger than 80% For the remain of attacks such as salt and pepper noise (density 1%) and gaussian 5%, the restored watermark images are up to 93% and 89%, respectively 20 (a) Left watermarked image Img1, Cropping 6,25%, PSNR=35.90 (b) Right watermarked image Img1, Cropping 6,25%, PSNR=35.97 (c) extracted watermarking image BCR=0.83 (d) Left watermarked image Img1, Cropping 25%, PSNR=29.99 (e) Right watermarked image Img1, Cropping 25%, PSNR=30.01 (f) extracted watermarking image BCR=0.81 (g) Left watermarked (h) Right watermarked Img1, salt&pepper 1%, Img1, salt&pepper 1%, PSNR=36.75 PSNR=36.91 (i) extracted watermarking image BCR=0.93 (j) Left watermarked image Img1,Gaussian 5%, PSNR=31.48 (k) Left watermarked image Img1,Gaussian 5%, PSNR=31.47 (l) extracted watermarking image BCR=0.89 Fig 4.1 PSNRs of attacked watermarked images and BCRs of extracted logo under various types of attacks 21 4.3 Secure and Robust watermarking scheme based on DCT-SVD and Chaotic logistic map encoding 4.3.1 Related works 4.3.2 Proposed scheme The proposed scheme can be illustrated in Fig 4.2 Which covers two processes: watermarking embedding and watermarking extracting Fig 4.2 The flowchart of the proposed scheme based on DCT-SVD and Chaotic logistic map encoding 4.3.3 Performance evaluations of the proposed scheme 4.3.3.1 Sensitive key analysis It is known that a chaotic logistic map is highly sensitive to every initial condition even with minute changes for an image encryption proposal, which guarantees to withstand a brute-force attack To evaluate the key sensitivity, we, firstly, scramble the plain image with two initial values x0 and  as 0.7589 and 3.67, respectively Subsequently, we change the initial value x0 by adding 10-10 to value x0 After that, we rearrange the cipher image Fig 4.3 displays the results of the key sensitivity 22 (a) Left view of stereo image Art (b) Left cipher image (c) decoding (e) Right view of stereo image Art (f) Right cipher image (g) Correct decoding Correct (d) decoding with x0+10-10 (h) decoding with x0+10-10 Fig 4.3 The results of the key sensitivity analysis 4.3.3.2 Adjacent pixels correlation analysis In order to test the security of the system, the pixels correlation analysis between the plain image and the cipher image is conducted We randomly select 2048 pairs of adjacent pixels in a horizontal, vertical and diagonal direction from the plain image as well as the cipher image The correlation of adjacent pixels can be visually analyzed by plotting Fig 4.4 displays the correlation distributions of the plain stereo image ‘Art’ with the red component before and after encoding Plain image (a) Horizontal (b) Vertical (c) Diagonal Cipher image (d) Horizontal (e) Vertical (f) Diagonal 23 Fig 4.4 Correlation visualization for 2048 randomly selected adjacent in plain and cipher images 4.3.3.3 Imperceptibility and robustness evaluation of the watermarking system To evaluate the watermark imperceptibility and robustness of the system, a logo watermark sized of 128×128 and stereo image dataset from Middlebury Fig 4.5 displays a logo watermark and some of the test stereo images (a) logo watermark (b) Left view of Laundry (c) Left view of Art Fig 4.5 Logo watermark and left view of test stereo images Table 4.1 gives the values of peak signal-to-noise (PSNR) and structural similarity (SSIM) index for test images The PSNR (dB) of the watermarked stereo images is about 43 for W Zhou et al.’s scheme5 While the PSNR of the watermarked stereo images of the proposed scheme is larger than 56 (dB), which is satisfied with watermark transparency Table 4.1 PSNR/SSIM of watermarked images W Zhou et al.’s scheme Proposed scheme Left view Right view Left view Right view Laundry 42.80/0.978 42.79/0.977 56.99/0.999 59.39/0.999 Art 43.20/0.987 43.10/0.980 57.24/0.999 59.85/0.999 Stereo images 4.4 Conclusion Chapter proposed two blind watermarking schemes that enhance the robustness and security of the system by protecting the watermark image and embedding positions The performance evaluations demonstrate that the vision.middlebury.edu/stereo/data, http://vision.middlebury.edu/stereo/data/ Zhou, W., Jiang, G., Luo, T., Yu, M., Shao, F., Peng, Z.: Stereoscopic image tamper detection and self-recovery using hierarchical detection and stereoscopic matching JEI 23, 023022 (2014) 24 proposed scheme meets a good imperceptibility of each view in comparison with the similar state-of-the-art algorithm The PSNR, SSIM metrics verified the better performance of the proposed watermarking scheme CHAPTER 5: CONCLUSION AND FUTURE WORKS 5.1 Conclusion Data hiding is becoming one of the most rapidly advancing techniques the field of research especially with an increase in technological advancements in internet and multimedia technology In this dissertation, novel data hiding techniques are proposed to embed secret data into Stereo images Enhancing the embedding capacity and efficiency of data hiding in digital images lie on the factors often termed as optimality criteria or constraints for data hiding The experimental results are presented and discussed in Chapter and Chapter 5.2 Future works The challenge of a hidden information system is that the more secret information is embedded, the more distortion will cause in the stego-images, leading to low visual image quality The research has achieved the goal of the dissertation enhancing the embedding capacity and efficiency of data hiding in Stereo images Nowadays, three-dimensional images have been used in industrial, medical and entertainment applications Thus, this is a good approach in data hiding technique based on 3-D images 25 ... DOCTORAL THESIS [CT1] Nguyễn Thái Sơn, Võ Phước Hưng, Huỳnh Văn Thanh, Đỗ Thanh Nghị, ? ?Giấu tin thuận nghịch ảnh Stereo với khả nhúng tin cao? ??, Kỉ yếu hội thảo FAIR 2016, pp 631-637 [CT2] P-H Vo, T-S... stegoimage, is still in a meaningful format Therefore, data hiding techniques can avoid attracting attackers’ attention 1.2 Data hiding mechanisms in digital images Along with powerful software, new... information is being exchanged which has attracted the attention of many unauthorized users such as snooping, masquerading or modification Therefore, protecting the security of such information is