Robust track-and-trace video watermarking

With the development of computers and the Internet, digital multimedia can be distributed and pirated easily. Watermarking is a useful technique for multimedia copyright protection. In this paper, we develop a robust video watermarking system. It consists of two components, i.e., watermarking embedder and watermarking detector. In the embedder, we insert a watermark pattern into video frames according to a watermark payload. The watermark pattern is generated from a Pseudo-random Noise (PN) sequence generator using the spread spectrum technique. User and copyright information are mapped to a binary sequence and then encrypted with Advance Encryption Standard (AES) and encoded/protected by convolutional error correction code (ECC) to produce watermark payload. The watermark pattern is weighted and embedded to each frame to meet perceptual requirements. In addition, the video is slightly geometrically manipulated in order to defend possible collusion attacks. The detector extracts the watermark from the candidate video. Kanade-Lucas-Tomasi feature tracker (KLT) is used to register the candidate video with respect to the original video, to enhance the correlation with the reference. The cross-correlation sequence is binarized, ECC decoded and decrypted. The experimental results show that the proposed video watermark system is very robust to not only geometric attack, but also collusion attacks, and that it is perceptually invisible to human vision system.