|M.Sc Student||Ginzburg Mark|
|Subject||Holographic Video-Scene Watermarking in the|
|Department||Department of Computer Science||Supervisor||Professor Alfred Bruckstein|
|Full Thesis text|
Watermarking is a process of embedding hidden information in a host signal. This work focuses on invisible and robust watermarking schemes for video sequences. A common requirement from the video watermarking schemes is the possibility of blind recovery, i.e. decoding with no access to the original video.
Applications of robust watermarking are various, like authorship rights protections, distribution monitoring and steganography. In this work we stress the importance of embedding the watermark in such a way that it will be holographic, i.e. it will have no spatial locality. The aim is to make watermarking robust against cropping, frame dropping and other common signal processing operations, hence part of the information corresponding to each “watermark bit” must reside in every big enough portion of video sequence. To avoid problems of temporal synchronization and efficiency, partition of the video into natural scenes is necessary.
In this work we discuss practical schemes for embedding watermarks in the 3D discrete Fourier transform of video scenes. First traditional Spread Spectrum scheme is introduced. Then it is shown how to improve this popular scheme using Turbo error correction codes and informed embedding.
Finally, a novel video watermarking scheme, based on modified trellis codes is proposed. It is demonstrated that combining this technique of informed coding with a Reed-Solomon error correction code generates a very robust scheme. The scheme is shown to be robust against common signal processing operations that do not destroy the quality of the host video. In addition it is shown that performance of this watermarking scheme is better then performance of other discussed schemes.