关键词: 尺度不变特征变换, 特征点匹配, 像素差平方和, 最小二乘法, 游程编码, 几何攻击

Abstract: It might fail to detect watermarking now that geometric attacks perform on a watermarked image and destroy the synchronization of the watermark signals embedded in the image. To restore the lost synchronism, a novel method to estimate the geometric operation using sifted Scale-Invariant Feature Transform (SIFT) was proposed. According to Sum of Squared Differences (SSD) of the pixels, the suitable feature points were refined to ensure robustness of the least-squares solution. The least squares iterative was convergent quickly, and the one step iteration would be able to get high accuracy geometric parameters; encoding binary watermark image by using Run-Length Code (RLC), watermarking was done by altering the coefficients of the DCT blocks corresponding to the value of RLC. The simulation results show that the proposed scheme achieves good image quality and it is robust to geometric attacks.

Key words: Scale Invariant Feature Transform (SIFT), feature points matching, Sum of Squared Differences (SSD), least-squares solution, Run-Length Code (RLC), geometric attack