Security Evaluation of Generalized Patchwork Algorithm from Cryptanalytic Viewpoint

Abstract

Abstract. In this paper we present a cryptanalysis of the generalized patchwork algorithm under the assumption that the attacker possesses only a single copy of the watermarked image. In the scheme, watermark is inserted by modifying randomly chosen DCT values in each block of the original image. Towards the attack we first fit low degree polynomials (which minimize the mean square error) on the data available from each block of the watermarked content. Then we replace the corresponding DCT data of the attacked image by the available data from the polynomials to construct an attacked image. The technique nullifies the modification achieved during watermark embedding. Experimental results show that recovery of the watermark becomes difficult after the attack.

Robustness Analysis of Patchwork Watermarking Schemes

“…The authors (Das, 2005) present the experimental results listed in Table 1, for three different 512 512 ×…”

“…Here we survey a cryptanalytic method of the generalized patchwork algorithm under the assumption that the attacker possesses only a single copy of the watermarked image (Das, 2005). In the scheme, the watermark is inserted by modifying randomly chosen DCT values in each block of the original image, referred to (Yeo, 2003b).…”

“…Consider a specific DCT block and now the authors (Das, 2005) explain the case 0 1 = P and 0 2 > P , i.e., 0 = β . If 0 has been inserted, then the values in DCT coefficients 0 A and B 0 will not change, but A 1 and B 1 will change.…”

“…It has been concluded that the MPA is sufficiently robust and inaudible. Recently, Das et al (2005) have performed cryptanalysis against the GPA for image watermarking based on the curve-fitting attack (Das, 2004) under the assumption that an attacker can get the information of how to segment image blocks. Note that this cryptanalysis against audio watermarking has not been performed yet.…”

“…The first method (Das, 2004) was verified effective to some watermarking techniques, but almost ineffective to the patchwork algorithm. The second approach (Das, 2005) transforms the time-domain samples, and applies curve-fitting to the unsorted transformed coefficients. It is noted that the DCT data remains unsorted to disturb more for the image patchwork algorithm as long as the information of image block size is leaked.…”