Motion Sensitive Video Watermarking preliminary study

The aim of this report is to provide background information on all topics in literature that are relevant to the development of a video watermarking method that will incorporate motion information in the embedding algorithm. First, a short introduction to digital video and digital image processing is given. Next, the following subjects are covered in more detail: video watermarking, the human visual system, video quality metrics and motion estimation. A general introduction to each subject is presented, followed by a discussion on several selected issues. Among these issues are: the working of the Philips watermark embedding and detection algorithm, the sensitivity to noise of the human visual system, subjective and objective metrics for image and video quality, the principles of motion estimation and a comparison of several motion estimators. Finally, specific issues regarding motion sensitive video watermarking algorithms are addressed, and a number of conclusions based on the preliminary literature study are discussed. Conclusions: This report is a survey of the information currently available in literature on the topics mentioned above. Nevertheless, some early conclusions can be drawn: Optimal transparent watermark embedding can only be achieved when a watermark embedding algorithm includes a sophisticated model of the human visual system. Due to performance constraints, only simple approximations of the human visual system have been incorporated into current watermarking technologies. Visual transparency of watermarks can be improved by accounting for more complex properties of the human visual system such as contrast masking and motion perception. Current perceptual quality metrics do not offer a reliable measure of video quality. They fall short in their representation of the characteristics of the human visual system. Hence, subjective tests remain the only true indicators of video quality. The class of motion estimators used for video compression is less suitable for a motion sensitive watermarking algorithm. These estimators do not reliably reflect the true motion of objects in a video sequence. The Y-prediction block-matcher, used in video de-interlacing applications, is light-weight and returns true-motion fields. It will be the first candidate to extend the current watermarking algorithm. c ©Philips Electronics N.V. 2001 iii 2001/824 no classification iv c ©Philips Electronics N.V. 2001