A content dependent spatially localized video watermark for resistance to collusion and interpolation attacks

This paper presents a novel video watermarking algorithm based on two key ideas: statistical invisibility and contentsynchronized placement. We argue that statistical invisibility is essential to protect video watermarks from statistical collusion, and present a natural way to induce this property using a content-dependent spatially localized watermarking framework. We introduce the notion of a watermark footprint, the spatial locations over which its energy is spread. By defining localized footprints with regular structures, e.g., a set of subframes within each frame, current image watermarking techniques can immediately be applied at the subframe-level. We address the issue of reduced spatial redundancy by proposing an attack model based on bilinear interpolation, and embedding the watermark into regions with lower expected distortions. Results are presented to demonstrate the effectiveness of the algorithm.