A Content-Dependent Video Watermarking Scheme For Anti-Collusion Attacks

After discussed the application, classification, characteristics, challenges and development trends of video watermarking technologies, in this thesis, we analyzed the collusion attacks of video watermarking, illustrated the properties and conditions for robust watermarking against collusion. The embedded watermarks can be regarded as high frequency noises, if they do not contain the content of video frames. Denoising operations may easily remove the watermarks from the video without damaging its quality. Therefore, we believed that it is an effective way for resisting collusion attacks through combining the watermarks with the content of video before embedding.In order to represent the content of video frames, we analyzed the basic requirements and the structure of image hashing system, and the recent development of image hashing technologies. Since the operations, such as image enhancement, geometric modification, and compression coding, do not change the essential content of the image, the image hashing, different from the traditional cryptographic hash functions used for content authentication, should be able to capture the essential content of the image, and be robust to changes insignificantly to the human eyes made to the image. The encoding methods used in current video compression standards usually divide the image into 8×8 pixel blocks and conduct DCT transformation on these blocks. In order to combine well with the video compression standards and reduce the complexity of the algorithm, we proposed a new image-hashing algorithm based on DCT domain in the thesis. The results of the experiments showed that the proposed image hashing algorithms enjoyed a good robustness against the operations such as filtering and cropping.After obtained the hash sequences that represented the video frames, a function such as permutation function was used to mix the hash sequences with the embedding watermark to produce a content-dependent watermark (CDW) that would be embedded into video frames using appropriate embedding algorithm.