Abstract: Rotation, scaling and translation (RST) attacks can desynchronize watermark detection, thus making many watermarking systems fail. In this paper, an image adaptive RST invariant watermark (AWPZM) is proposed by using the rotation invariant property of pseudo-Zernike moments and odd-even quantization. Pseudo-Zernike moments of the original image are computed firstly, and those suitable for watermark generation are selected. Then, their magnitudes are odd-even quantized to generate the watermark. In detection, the normalized hamming function is employed to weigh the similarity of the watermark. Experimental results show its robustness to rotation and scaling. For traditional attacks such as JPEG compression, added noise and filtering, the similarities are all above 0．95.

Key words: information hiding, adaptive watermark, pseudo-Zernike moments, odd-even quantization