A new watermarking attack based on content-aware image resizing

In this paper, we propose a new method for damaging and destroying robust invisible watermarks using an image resizing technique which is named seam carving. By using this method we are able to resize watermarked images in a content-aware manner so that the synchronization of the embedder and extractor of watermarking system is broken and the watermark detection becomes impossible. In contrast to the available benchmarks like Stirmark, proposed attack does not severely reduce the quality of the watermarked image. Therefore it maintains the commercial value of the watermarked image. We have tested the proposed method to attack 3 recent and robust watermarking methods and the results sound impressive. The NC for all extracted watermarks after applying this attack is lowers than predefined threshold 0.4, so they are not detectable. The proposed method is a generic attack which does not consider any knowledge about the underlying watermarking algorithm.