Collusion Attack Resistant Watermarking Scheme for Colored Images using DCT Vikas Saxena, J.P Gupta Abstract- Image watermarking with both insensible detection and high robustness capabilities is still a challenging problem for copyright protection up to now. This paper presents a new scheme for hiding a logo-based watermark in colored still image which is inherently collusion attack resistant. This scheme is based on averaging of middle frequency coefficients of block Discrete Cosine Transform (DCT) coefficients of an image. It is different from earlier schemes based on middle frequency coefficient by mean of high redundancy, to sustain malicious attacks. Experimental results show the robustness of the proposed scheme against the JPEG compression and other common image manipulations. Index Terms- Collusion attack, Discrete Cosine Transform (DCT), Image watermarking, JPEG compression. I. INTRODUCTION 1 With digital multimedia distribution over World Wide Web, authentications are more threatened than ever due to the possibility of unlimited copying. So, watermarking techniques are proposed for copyright protection or authentication of digital media. Many watermarking methods for images have been proposed [1]- [4]. More and more researchers are joining this area and number of publications is increasing exponentially. Most of the work is based on ideas known from spread spectrum communication [5] which is additive embedding a pseudo- noise watermark pattern and watermark recovery by correlation [6]. Cox et al suggested using the DCT domain [6], which has been extensively studied because this is the transform used in JPEG compression. Further advantage of using DCT domain includes the fact that frequency transform is widely used in image and video compression and DCT coefficients affected by compression are well known. This paper proposes an efficient use of middle-band coefficients exchange to hide the watermark data. This paper uses the idea of Middle Band Coefficient Exchange which was Manuscript received February 15, 2007. This work is a part of a doctoral degree under the Jaypee Institute of Information Technology University, Noida, India Vikas Saxena is Senior Lecturer in CSE & IT Department Jaypee Institute of Information Technology University, Noida, India (phone: +91-120-2400974- 326; fax: +91-120-2400976; e-mail: vikas. saxena@jiit. ac. in) J. P. Gupta is vice chancellor of Jaypee Institute of Information Technology University, Noida, India (e-mail: jp. gupta@jiit. ac. in) discussed by Koch and Zhao [8] and further explained by Johnson and Katezenbeisser [9]. Later Hsu and Wu also used the DCT based algorithm to implement the middle band embedding [10]. Further one more efficient collusion attack resistant scheme has been presented based on middle-band coefficients exchange [42]. Collusion attack is the severe problem for some applications of watermarking like fingerprinting which involve high financial implications. So while designing a watermark scheme we are taking this attack as a prime. [43]- [45] Our main motivation behind selecting middle-band coefficients exchange scheme as a base is that this scheme has proven its robustness against those attacks which any how do not affect the perceptual quality of an image such as JPEG compression. Section 2 discusses the background studies. Section 3 describes the proposed method and section 4 discusses the results. II. PRELEMINARIES Classical Middle-band based algorithm interchanges only one pair of coefficients and is quite robust against JPEG compression and common image manipulation operations but vulnerable to collusion attack. A. Middle-band Coefficient Exchange Scheme The middle-band frequencies coefficients (F M ) of an 8x8 DCT block are shown in Figure 1. F L is used to denote the lower frequency coefficients of the block, while F H is used to denote the higher frequency coefficients. F M is chosen as embedding region to provide additional resistance to lossy compression techniques, while avoiding significant modification of the cover image. First we take 8x8 DCT of original image. Then two locations DCT (u 1 , v 1 ) and DCT (u 2 , v 2 ) are chosen from the F M region for comparison of each 8x8 block. We should select the coefficients based on the recommended JPEG quantization table shown as Table-I. If two locations are chosen such that they have identical quantization values in JPEG quantization table, then any scaling of one coefficient will scale the other by the same factor to preserve their relative strength. Based on Table-I, we observe those coefficients at location (4, 1) and (3, 2) or (1, 2) and (3, 0) are more suitable candidates for comparison because their quantization values are equal. The DCT block will encode a “1” if DCT (u 1 , v1) > DCT (u 2 , v2); otherwise it will encode a “0”. IAENG International Journal of Computer Science, 34:2, IJCS_34_2_02 ______________________________________________________________________________________ (Advance online publication: 17 November 2007)