Due to the possibility of easy unauthorized reproduction of digital data and the explosive expansion of Internet in the last decade, the copyright protection of digital data has become a very important issue. To solve this problem, digital image watermarking techniques have been proposed for copyright protection of digital images recently. And the new standards such as MPEG-4 and JPEG2000 offer an object-based representation of image, where arbitrarily shaped image objects can be coded into separate bit streams. However, there is no watermarking method that can be applied to arbitrarily shaped image objects. Also the conventional watermarking methods that do not require the original image are most vulnerable to geometrical attacks such as scaling and rotation. In order to overcome these problems, we propose two new image watermarking schemes for arbitrarily shaped image objects and common rectangular image, respectively.
The object-oriented watermarking method can be applied to any arbitrarily shaped image object. It is designed to be robust against the geometrical attacks if there is no variation of the shape of the object. Using coordinates conversion to log-polar and a new shape adaptive discrete transform(PSADT:Polar coordinates Shape Adaptive Discrete Transform), the proposed watermarking method extracts the coefficients that are invariant to scaling and rotation and then hides an invisible signal, known as watermark, in them. Experimental results demonstrate that the proposed method is much robuster against geometrical attacks such as rotation and scaling than the conventional methods.
In order to overcome the problem of the conventional watermarking methods against geometrical attacks and to apply the object-oriented watermarking method to a common rectangular image, we propose a new watermarking scheme named as two-layer watermarking method. It consists of a marking layer and a synchronization layer for the specialized defense against two kinds of attacks, i.e., waveform attack and synchronization attack. Namely, the synchronization layer only determines the target region where the watermark signal will be embedded or extracted, and the marking layer spreads the watermark signal over the image as the conventional watermarking methods do. For the design of robust synchronization layer, we propose a framework using inherent image structure. The proposed synchronization layer is made up of three procedures, i.e, image simplification, inherent image structure analysis and target region extraction. Specifically the detailed signal components that are unnecessary to analyze the image are removed in the simplification procedure. And the salient points that reflect the structure of the simplified image are determined through the analysis procedure. To quantify the structure in the analysis procedure, we take advantage of local intensity distribution or magnitude of corner. Finally based on the location of the salient points, the target region extraction procedure forms the region where the watermark signal will be embedded or extracted. Also we present a specific way to embed the watermark based on the object-oriented watermarking method and propose a search scheme for more robust extraction of the watermark at the marking layer. We demonstrate the proposed synchronization layer forms consistent target region and the two-layer watermarking method has outstanding performance over the conventional watermarking methods against various attacks.