Digital Image Watermarking: On the Enhancement of Detector Capabilities

Digital watermarking seems to be one of the most promising methods for copyright protection of digital images, authentication, labeling and monitoring. Since digital images can be published in the Internet, anyone can access and manipulate them. Such kind of manipulations (attacks) falls in one of the following categories: (a) those applied without malicious purposes and (b) those applied with the aim of removing or destroying the watermark. Our model recovers geometric transformations (i.e. rotation, scaling, translation or a combination of them). Geometric transformations are considered as some of the basic attacks in digital watermarking in both of the above-mentioned cases. The model focuses not on the watermarking method, but rather on the detector’s capabilities. This is achieved by selecting several regions in an attacked watermarked image, which hold sufficient information about the watermark, and trying to apply an inverse operation in order to extract the watermark. It is assumed that one or more watermarks are spread in the image. An analysis of the attack methods has given the information about the effectiveness of every function or combination of functions in the watermarked material, the amount of data needed in order to represent such an operation and the possibility of an operation to be inverted. By means of all this information we can give a solution to the blind image registration problem in watermarking of digital images, i.e. the process of inverting the distortion of the attack with the aim to detect the watermark.