Resilient video transmission and multimedia database applications

Video transmission is subject to imperfections of real channels, including insufficient and dynamically varying bandwidth. The problem is particularly serious in the case of multipoint communications over heterogenous networks. Current coders are designed without giving much consideration on the effect of such imperfections.; A structure-based directional interpolation scheme with minimal computation will be presented for the recovery of lost blocks due to channel noise or network congestion. This structure-preserving approach usually can recover high frequency details faithfully, as opposed to other techniques. A more proactive approach for resilient video transmission is to first estimate the instantaneous available bandwidth in a given network, and then adaptively shape the bit rate of video in the compressed domain down to match to the available bandwidth. A compressed-domain rate shaping approach will be described to reduce the bit rate of MPEG video, by intentionally dropping selective blocks, in order to match the instantaneous bandwidth. A joint block dropping and coefficient truncation approach will also be presented. Furthermore, a method will be presented to protect perceptually salient features during rate shaping by pre-analyzing video so as to produce visually superior image quality. The general idea of intelligent data dropping combined with subsequent recovery also finds applications in low bit rate coding, unequal error protection, etc..; An important issue related to video transmission is the analysis of the video data for later efficient transmission and easy retrieval. The other theme of the thesis is to address some issues in multimedia database applications, including content-based video analysis and digital image watermarking. It will be shown that incorporating other media data into the video indexing system will increase the indexing accuracy and reduce the overall computation. Media security becomes an important issue due to increasingly easy access to digital media and increasingly powerful digital media manipulation tools. The final part of the thesis addresses digital image watermarking. The advantages of incorporating visual models to achieve the tradeoff between perceptual invisibility and robustness of invisible watermarks are demonstrated. The particular applications of digital watermarking for resolving rightful ownership and feature/content-based video manipulation are also studied.