Geometrical Distortion-Resilient Watermarking Based on Image Features

The major threat of geometric manipulations is that they change the positions of watermarks, therefore the detection process fails to extract watermark properly. Since they cause the same effects on the host image as watermarks simultaneously, evaluating the distorted host image can be helpful to measure the nature of distortions. In this paper, we propose a geometrical distortion-resilient watermarking algorithm based on this property. Firstly we evaluate the orientation of a host image by filtering it with directional Gabor kernels, then we insert embedding pattern aligned to the estimated orientation. In its detection step, we evaluate the orientation again by Gabor filtering, then simply project and average the projected value to obtain a 1-D projection average pattern. Finally, auto-correlation function of the 1-D projection average pattern identifies periodic peaks. Analysed are experimental results against geometrical attacks including aspect ratio changes.