| Summary: | Recently, watermarking has been modelled as communications with side information at the transmitter. The advantage of this is that in theory the interference due to the cover Work or host signal can be eliminated, thereby improving the capacity of the watermarking system. Hence a number of different practical methods have been proposed, one of which is based on dirty paper trellis coding. These codes are a form of spherical code, and as such, have the advantage of being robust to amplitude scaling. Dirty paper trellises have a number of design parameters. There is a lack of understanding on the influence of these parameters on performance, and this thesis attempts to address this. In particular, the thesis examines the following parameters: (i) the number of states and the number of arcs per state in the trellis, (ii) the distribution of the codewords generated by the trellis, and (iii) the cost function associated with each arc. Experimental results are provided on both synthetic signals and real images that demonstrate how performance is affected and a number of suggestions and improved designs are discussed. In particular, a deeper understanding of trellis configurations is provided that serves as a foundation on which to choose the best trellis structure based on bit error rate performance and computational cost. Secondly, trellis coded modulation (TCM) is adapted for use in a dirty paper trellis. This results in an improved distribution of the codewords on the sphere which leads to improved performance. Lastly, during embedding, the embedder usually searches for the codeword that has the highest linear correlation with the cover Work. However, this codeword may be difficult to embed due to perceptual constraints. We show that searching for a codeword that maximises a cost function based on linear correlation and perceptual distance can significantly improve performance.
|
|---|
