<title>Review on attacks, problems, and weaknesses of digital watermarking and the pixel reallocation attack</title>

Abstract

Watermark attacks are first categorized and explained with examples in this paper. We then propose a new image watermark attack called "Pixel Reallocation Attack". The proposed attack is a hybrid approach, which aims to decorrelate the embedded watermark with the original watermark. Since many watermarking detections are by correlating the testing image with the target watermark, it will not work once we decorrelate the embedded watermark. For example, the geometrical transformation attacks desynchronize the correlation detector with the testing image leading to detection failure. However, by inserting a template or grid into the watermarked image can make inverse transformation possible and the watermark can be retrieved. If we apply transformations to every single pixel locally, independently and randomly, inverse transformation will not be possible and the attack will be successful since the embedded watermark is not correlated with the original watermark. Experiment shows that single technique approach needs a larger distortion to the image in order to attack the image successfully. We also tested our attack with commercial watermarking software. It cannot detect the watermark after we applied the proposed hybrid attack to the watermarked image.

Robust Watermarking Framework for High Dynamic Range Images Against Tone-Mapping Attacks

“…Image watermarking is a common method of proving ownership or determining origin (Tsang & Au, 2001). Unintentionally destroyed watermarks happen when transmitting an image.…”

Still-image watermarking robust to local geometric distortions

“…The watermark cover is divided into blocks whose size is pixels. We define a block of as follows: (8) where are the indices of the block, are the coordinates of the displacement and represent the coordinates of an element of the block.…”

“…Until now, most of the proposed watermarking algorithms, derived from classical schemes such as [3]- [6], are mainly robust against usual image processing such as low-pass filtering or JPEG compression. However, most of them are still weak against geometrical distortions and malicious attacks [7], [8] (e.g., Stirmark [9], [10], collusion attacks, Dewatermarking Attack [11], etc.). It is quite easy to see that data payload, fidelity, and robustness are often in conflict.…”