Image Authentication and Recovery Scheme Based on Watermarking Technique KENJI SUMITOMO 1, MARIKO NAKANO 2, HECTOR PEREZ 2 1 Faculty of Information and Computer Engineering The University of Electro-Communications Chofugaoka 1-5-1, Chofu City, Tokyo JAPAN 2 Graduate Section of ESIME Culhuacan National Polytechnic Institute Av. Santa Ana no. 1000, Mexico D.F. MEXICO hmpm@prodigy.net.mx Abstract:- Actually digital images are used to show some important evidences, however once they are propagated through an open channel such as Internet, these images are easy targets of malicious modification. In this paper, we propose a watermarking based image authentication scheme, in which owner s blocks of interest are authenticated using embedded watermark. If the scheme determined that these blocks were modified intentionally, the resume of the original version of these blocks is recovered without any additional information. Also the embedded watermarks are enough robust to no-intentional attacks, such as JPEG compression and noise contamination. In the watermark embedding process, watermark sequence is extracted from the blocks of interest of the image and it is embedded into the corresponding DCT blocks indicated by a previously generated mapping list. In the authentication and recovery process, the watermark sequence is extracted from the corresponding DCT blocks and then it is compared with the sequence extracted from the blocks of interest to determine its authenticity. The proposed system use only a secret key to generate the mapping list to map owner s blocks of interest and other blocks. The evaluation results show efficient authentication and recovery capacity of the proposed scheme. Key-Words:- Image Authentication, Recovery, Watermarking, DCT, JPEG compression, Blocks of Interest 1. Introduction Actually digital images are used to show some important evidences and events, however once these images are propagated through an open channel such as Internet, non authorized persons can edit easily them, changing some evidences of the image in an imperceptible manner. Therefore the digital image authentication becomes more and more important issue. There are three types of authentication: complete authentication, robust authentication and content authentication. Complete authentication determines its authenticity if the image under evaluation is identical to its original version, robust authentication determines if the image under evaluation is its original version or it s compressed one, or if the image under analysis corresponds to the original one with a small degradation within the visually permitted level; and finally content authentication is content preserve authentication, which determine its authenticity while the content of image is unchanged. Basically digital image authentication schemes can be classified by two categories, which are digital signature based scheme [1-3] and watermarking based scheme [4-9]. The digital signature based scheme extract bits sequence from the image and it is transmitted together with the image via communication channel, while watermarking based scheme the extracted bit sequence is embedded into the same image in imperceptible manner. Here only watermarked image is transmitted via communication channel. Generally digital signature based scheme has disadvantage compared with the watermarking based scheme, because former scheme increase total bits to be transmitted via communication channel to do authentication. In both schemes, depending on the extracted bits sequence from the image, three types of authenticator (complete, robust and content) can be constructed. ISSN: 1790-5117 Page 94 ISBN: 978-960-6766-33-6
The watermarking based image authentication scheme using robust bits sequence is called semifragile watermarking scheme [4,5], which detect malicious modification while tolerate to some non intentional modifications, such as compression and contamination by noise, etc. On the other hand, watermarking scheme realizes complete authentication is called fragile watermarking scheme [6,7]. Many image authentication schemes only determine integrity of the image and detect regions suffered some modifications, however a few schemes have recovery capacity of modified regions [8-10]. In the authentication schemes proposed by [8] and [9], the image is divided in 8x8 sub-blocks and mapping list between sub-blocks is generated using the chaotic mixing method with a secret key. In [8], watermark bits sequence is formed by a resume of each block of image, output bits of hash function of its resume and a cyclic redundancy check bits. While in [9], the watermark bits sequence is formed by a resume of block of image, an authentication bit and a parity bit. In both scheme, the watermark bits sequence embeds into two LSB s of the corresponded image block. From the embedding method and the extracted watermark sequence, both schemes are classified as fragile watermarking scheme and embedded watermark sequence is not robust to no-intentional modifications, such as image compression, contamination by noise, etc. In [10], authors proposed image authentication scheme with recovery capacity of modified area, in which the watermark sequence is embedded into SPIHT encoded list of significant pixels bit stream. This scheme can be robust to no-intentional modification, although the robustness is not shown by authors; and its hiding capacity is limited. This paper proposes a watermarking based image authentication scheme, in which image contents of the blocks of interest are authenticated using embedded watermark. If the proposed authenticator determined these blocks are modified intentionally, the resumes of the original version of these modified blocks are recovered without any additional information. The rest of this paper is organized as follows. In Section 2, the proposed scheme is described in detail, and the experimental results will be shown in Section 3. Finally some conclusions will be given in Section 4. 2. The Proposed Authentication Scheme The proposed authentication scheme consists of watermark generation and its embedding stage, and image authentication and recovery stage. 2.1 Watermark embedding stage The watermark generation and embedding stage is shown by Fig. 1. Fig. 1 Watermark generation and embedding stage Watermark generation and embedding stage is consisted of following steps: 1. Selection of blocks of interest Some blocks of interest (8x8 pixels) are selected, which are important region of the image that requires a protection against malicious modification. 2. Mapping list generation Using a secret key, mapping list between blocks of interest and other blocks of the image is constructed. 3. Extraction of watermark bits sequence From each block of interest, 96 bits of resume of the image is extracted, which is used as watermark bits sequence. To extract this bits sequence, each block is divided into sub-blocks ISSN: 1790-5117 Page 95 ISBN: 978-960-6766-33-6
of 2x2 pixels and we get average of these 4 pixels values of each sub-block. First 6 MSBs of averages of all sub-blocks of a block of interest consists a sequence of 96 bits. The 96 bits sequence is divided by 8 vectors of 12 bits data. 4. Watermark embedding Using the mapping list generated in step 2, each block of interest is mapped into 8 blocks of the image to embed watermark sequence (total 96 bits extracted in previous step). The selected 8 blocks are transformed using 2D-DCT, and quantified by quantized matrix Q of predefined quality factor Qf as shown by (1). ( bits ) ( ) if XOR W, Ex < Th then the block is authentic if XOR W, Exbits Th then the block is modified (2) 4. Recovery If the authenticator determines that the block is modified intentionally, the recovery process is performed. This process replaces the block of interest by the extracted watermark sequence. Cuv (, ) = Cuv (, ) Quv (, ) (1) Where Cuv (, ) and Cuv (, ) are the (u,v)-th DCT coefficient and the quantized DCT coefficient, respectively, is lower nearest integer value of x. Each 12 bits data is embedded into the middle frequency band of each one of the 8 quantified DCT blocks. 5. Watermarked image generation Watermarked DCT blocks is multiplied by Q and transformed by inverse 2D-DCT to get watermarked blocks. Concatenating all blocks, we get watermarked image. 2.2 Authentication and recovery stage Authentication and recovery stage of the proposed scheme is shown by Fig. 2. This stage consists of following steps. 1. Mapping list construction The same operation of the embedding stage is performed. 2. Watermark sequence extraction 8 corresponded blocks for each block of interest are selected using mapping list. Each block is transformed by using the 2D-DCT and quantized by quantification matrix Q. From middle frequency band of each one of the 8 blocks, 12 bits sequence is extracted. For each block of interest, a total of 96 bits watermark sequence is extracted. 3. Authentication Watermark sequence W (96 bits) is compared with 96 bits sequence Ex bits extracted from block of interest and using threshold value, we determine authenticity of the block. Determination of authenticity is given by (2). Fig. 2 Authentication and recovery stage 3. Experimental Results The proposed authentication scheme is evaluated from various points of view: watermark imperceptibility in watermarked images with different number of blocks of interest, watermark robustness to no-intentional modification and capacity of authentication and recovery. 3.1 Watermark imperceptibility The figure 3 shows some watermarked images with a total of 6144 bits watermark (64 blocks of interest, 96 bits/ block). The quality distortions of four watermarked images respect to their original version are (a) 45.34dB, (b) 46.06dB, (c) 45.01dB and (d) 42.34dB, respectively. The watermark imperceptibility depends on total watermark bit length. According as increase the number of blocks ISSN: 1790-5117 Page 96 ISBN: 978-960-6766-33-6
of interest, the watermark bit length is increased and obviously the quality distortion of watermarked image is also increased. The figure 4 shows a relation between number of blocks of interest and PSNR (Peak Signal Noise Ratio). From the figure, the distortion of watermarked images with 8640 bits watermark sequence extracted from 90 blocks of interest is more than 40 db of PSNR, which means watermark is imperceptible by human visual system. 3.2 Watermark Robustness Fig.5 Robustness to JPEG compression (Qf=70). (a) Watermarked image, (b) Authentication results of Watermarked and compressed image and (c) Recovery blocks. Fig. 6 Robustness to JPEG compression (Qf=60). (a) Watermarked and compressed image by JPEG compression (Qf=60), (b) Authentication results and (c) Recovery blocks. Fig. 3 Watermarked images with 6048 watermark bits embedded. Watermark imperceptibility is (a) Barbara, 45.34dB, (b) Boat, 46.06dB, (c) Goldhill, 45.01dB and (d) Mandrill, 42.34dB. Fig.4 Relation between number of blocks of interest and PSNR. The proposed authentication scheme guarantee robustness to JPEG compression until a quality factor Qf= 70, because in the embedding process we use quantification matrix for Qf=70. If watermarked image is compressed by a quality factor Qf 70, the proposed scheme could determine authenticity correctly, however if the watermarked image is compressed by a quality factor Qf<70, authenticator could determine authenticity of blocks of interest erroneously and the distortion of recovery blocks is also increased. Figure 5 shows the robustness of the proposed scheme to JPEG compression, Fig. 5 (a), (b) show the watermarked image and its compressed version by JPEG compression with Qf=70. In this case authenticator determine authenticity of blocks of interest, therefore recovery process is not performed. Fig. 5 (c) shows recovery region, although it isn t replaced with the blocks of interest. Figure 6 shows robustness of the proposed scheme when watermarked image is compressed by quality factor Qf=60. Here Fig. 6 (a) shows watermarked and compressed image and Fig. 6 (b) shows the authentication results in detecting modified blocks, as indicated by black blocks, and Fig. 6 (c) shows recovery blocks. ISSN: 1790-5117 Page 97 ISBN: 978-960-6766-33-6
Also robustness to impulsive noise of the proposed scheme is evaluated. Figures 7 and 8 show noise corrupted watermarked images and its corresponded recovery blocks of interest. Impulsive noise density in Fig. 7 and Fig. 8 is 0.1% and 0.5%, respectively. The authenticator determines authenticity of blocks when the image corrupted by impulsive noise with density 0.1%, while when the image is contaminated by noise with 0.5% density, authenticator determines that the block is modified. changing the car number to L74047, and Fig. 9 (e) is recovery image constructed by extracted watermark sequence. Fig. 9(f) and (g) show zoomed image of blocks of interest before and after recovery, respectively. The distortion of recovery blocks compared with original ones is approximately 37.71 db. Fig. 7 Robustness to impulsive noise with density 0.1%,(a) Watermarked and noise corrupted image, (b) recovery blocks. Fig. 8 Robustness to impulsive noise with density 0.5% (a) Watermarked and noise corrupted image, (b) recovery blocks. 3.3 Authentication and Recovery Capacity Figure 9 shows authentication and recovery capacity of the proposed scheme. Fig. 9 (a) shows an original image, and Fig. 9 (b) shows the original image with blocks of interest (the region printed the car number L75047 ) indicated by black box area. Fig.9 (c) is the watermarked image using the bits sequence extracted from the block of interest as watermark sequence, taking a PSNR is 48.15dB respect to the original one. Fig. 9 (d) is malicious modified image, Fig. 9 Authentication and recovery result. (a) Original image, (b) original image with blocks of interest, (c) watermarked image, (d) maliciously modified image, (e) recovery image. (f) and (g) zoomed image of blocks of interest before and after recovery. 4. Conclusions In this paper, a watermarking based image authentication scheme is proposed. Watermark bits sequence is generated from selected blocks of interest and it is embedded into other blocks in the DCT domain. In the authentication stage, authenticator compares between extracted watermark sequences and extracted bits sequence of blocks to determine its authenticity. If authenticator ISSN: 1790-5117 Page 98 ISBN: 978-960-6766-33-6
determines that the block is modified maliciously, the block of interest is replaced by watermark sequence. Simulation results show that watermark imperceptibility and robustness to JPEG compression and contamination by impulsive noise. Also the recovery capacity of the proposed scheme, if authenticator determines that the blocks of interest were modified intentionally, was shown. Recovery, Int. Conf. of Information, Communication and Signal Processing, 2005, pp. 865-869. References: [1] C.-Y. Lin and S. -F. Chang, A Robust Image Authentication Method Distinguishing JPEG compression from Malicious Manipulation, IEEE Trans. Circuit Syst. Video Technol. vol. 11, no. 2, 2001, pp. 153-168. [2] C.-S Lu, H. -Y. Liao, Structural Digital Signature for Image Authentication: An Incidental Distortion Resistant Scheme, IEEE Trans. Multimedia, vol. 5, no. 2, 2003, pp.161-173. [3] D.-C. Lou, J. -L. Ju, Fault Resilient and Compression Tolerant Digital Signature for Image authentication, IEEE Trans. Consumer Electron. Vol. 46, no. 1, 2000, pp. 31-39. [4] Z. -M. Lu, D. -G. Xu, S. -H. Sun, Multipurpose Image Watermarking Algorithm Base don Multistage Vector Quantization, IEEE Trans. Image processing, vol. 14, no. 6, 2005, pp. 822-831. [5] K. Maeno, Q. Sun, S. -F. Chang, M. Suto, New Semi-Fragile Image Authentication Watermarking Techniques Using Random Bias and Nonuniform Quantization, IEEE Trans. Multimedia, vol. 8, no. 1, 2006, pp. 32-45. [6] P. -W. Wong, N. Memon, Secret and Public Key Image Watermarking Schemes for Image Authentication and Ownership Verification, IEEE Trans. Image processing. vol. 10, no. 10, 2001, pp. 1593-1601. [7] M. Celik, G. sharma, E. Saber, A. Tekalp, Hierarchical Watermarking for Secure Image Authentication With Localization, vol. 11, no. 6, 2002, pp. 585-595. [8] P. -L. Lin, P. -W. Huang, A. -W. Peng, A Fragile Watermarking Scheme for Image Authentication with Localization and Recovery, Proc. of the IEEE sixth Int. Symp. on Multimedia Software Engineering, 2004, pp.146-153. [9] P. -L. Lin, C. -K. Hsieh, P. -W. Huang, Hierarchical Watermarking Scheme for Image Authentication and Recovery, IEEE Int. Conf. on Multimedia and Expo, 2004, pp. 963-966. [10] P. Tsai, Y.-C. Hu, A Watermarking-Based Authentication with Malicious Detection and ISSN: 1790-5117 Page 99 ISBN: 978-960-6766-33-6