Improving Watermark Robustness by Combining Spatial and Frequency Domain Strategies

摘要

In this paper a novel approach, which combines frequency and spatial strategies to achieve a robust watermarking is presented. A frequency-domain technique is used to encode the watermark in the host image. This watermarking approach is robust against filtering, compression and most common image processing transformations apart from geometric distortions. The spatial domain is used to extract relevant information relating to the spatial distribution of pixels in the original image. The basic strategy consists of characterising salient image points using first order differential invariants. These primitives are invariant to orthogonal image transformations. To detect geometrical attacks in a frequency-domain watermarked image a robust matching technique is applied. Features extracted from the original image are matched with those extracted from the attacked image. Their disparities reveal the transformation parameters and are used to re-synchronize the attacked image. Experimental results confirm the robustness of the proposed schema.