<title>Robust compressed video watermarking</title>

Abstract

This paper investigates two methods for digital watermarking of MPEG2 video sequence. The watermark is spread by Frey's pseudo-chaotic sequence prior to embedding and it could be retrieved by a correlator from the watermarked video without the use of the original video. Both spatial- and DCT (Discrete Cosine Transform)-domain watermark are considered. They are embedded into every I or all I, P and B frames of the MPEG2 video. Some watermark parameters are adjusted to investigate the trade-off between watermark robustness and video quality or capacity of the watermarking system. The quality of watermark is measured by its BER (bit error rate) while the fidelity of watermarked video is measured by its PSNR (Peak Signal to Noise Ratio) as well as by visual inspection. Various forms of cropping attack are found to be detrimental to watermark retrieval but acceptable to the eyes in most cases. In particular, a new form of video watermark attack, i.e. frame cropping, is introduced. Their effects on the watermark robustness are studied and discussed.