Blind Attack on Watermarked Images Based on Segmentation and Non-Local Similarity

In this paper, a new attack is proposed on blind image watermarking, which is able to destroy the watermark embedded in an image using quantization and dither modulation embedding methods. This attack is consist of three main steps: The first two steps are segmentation and finding more similar regions of image and the third step is swapping similar regions with each other. One of the main idea in the proposed method is the algorithm of finding similar segments, which are far from each other and also are in indistinctive locations. This attack does not need to know the embedding and extraction algorithms and their parameters which are used. Therefore, this method is a blind attack. The results of this algorithm and its comparison with other intentional and unintentional attacks show that it can destroy embedded watermark properly and can preserve the quality of watermarked image. The value of NC metric is less than 0.4 and the value of PSNR metric is about 39 dB between the watermarked image and the attacked image.