A statistical approach for optimal watermark coefficients extraction in HVS-based blind watermarking system

In this paper, we make a study of the statistical approach for optimal watermark coefficients in HVS-based blind watermarking system and experiment in proposed TSDA (Two-Step Detection Algorithm) watermarking system. The watermark gain, alpha is optimized in robustness, invisibility, and capacity. We extract a watermark gain, alpha considering that BER is less than 20%, MPSNR is more than 43 dB (or PSNR is more than 38 dB), and C(capacity) is half of the full capacity of image. This optimal watermark gain, alpha, improve the performance of watermark systems. TSDA consists of two processes. In embedding, each bit is inserted in a PN spreading pattern in 12 positions of DCT coefficients of an 8times8 block. In detection process, a preliminary decision is obtained by correlation matching and then verified by choosing the more similar spreading pattern with the restored one. Most of bit errors are corrected in verification process. The proposed method has been tested for several test images, with attacks including lowpass/median filtering and JPEG compression, but excluding geometrical RST attacks. After verification, BER reduces to 0.5% with no attack. Even under heavy JPEG compression, BER stays lower than 9%. Compared with other methods, the proposed method - optimal watermark and TSDA - is better in watermark detection