Refinements in a DCT based non-uniform embedding watermarking scheme

Abstract

Perceptual watermarking is a promising technique towards the goal of producing invisible watermarks. It involves the integration of formal perceptual models in the watermarking process, with the purpose of determining those portions of an image that can better tolerate the distortion imposed by the embedding and ensuring that the watermarking will inflict the least possible degradation on the original image . In a previous study the Discrete Cosine Transform was used, and the watermark embedding was done in a non -uniform manner with criteria based on both the host image and the watermark. The decoder model employed made use of apriori access to unmarked and marked images as well as to the watermark. A fair level of success was achieved in this effort. In our research we refine this scheme by integrating a perceptual model and by proposing a modification to the decoder model that makes possible the successful recovery of the watermark without apriori access to it. The proposed perceptual scheme improves the watermark's transparency while at the same time maintains sufficient robustness to quantization and cropping. The proposed semi-blind variation offers adequate transparency and robustness to quantization, but its performance against cropping is considerably degraded.