| In this thesis, we propose two video watermarks based on the novel ideas of statistical invisibility and content-synchronized placement. We present a mathematical framework designed to facilitate collusion analysis and thereby enable protection from this important class of attacks. We define statistical invisibility and show that it is not only a property that supports the desired resistance to such attacks, but that it can also be easily induced using a spatially localized image-dependent approach. To construct the watermark, the notion of a watermark's footprint, the spatial locations over which its energy is spread, is introduced. By defining localized footprints with regular structures, e.g., sets of subframes within each frame, current image watermarks can immediately be applied at the subframe level. Results are presented to demonstrate the effectiveness of the algorithms. Comparisons are made with the well-known JAWS and CDMA video watermarks, as well as the StirMark 3.1 benchmarking suite. |
