Content based image watermarking using multi-level decomposition and CDMA technique

Watermarking algorithms robustness evaluation becomes recently a major preoccupation of a lot of researchers in image processing field. Recent works based on space domain decomposition use Delaunay tessellation according to the image content (1). The proposed decomposition gives a set of strongly correlated blocks with various sizes. To reach this goal we perform image decomposition using the Quadtree technique. The decomposition is based on statistical content analysis by using the Karhunen-Loeve Transform (KLT). The transform computes the energy concentrated on first eigenvectors to fix blocks homogeneity. The generated grid of blocks must be stable and must resist to a large number of image processing algorithms (2). The number of bits inserted in each bloc depends on bloc dimensions. In fact, largest blocks can contain more than one bit, one bit by layer, as proposed by CDMA approach (3). Besides, our approach uses a scalar coefficient, because of blocks homogeneity, to perform the psycho-visual watermark weighting and avoid computing this coefficient locally for each pixel, which is very time consuming.