Scalable subband/wavelet coding and space-frequency block matching for digital video

grantor: University of TorontoEfficient representation of digital video signals is key to many products and applications that will soon be extensively deployed. The scope of this thesis is the core technology for conventional compression of video: motion estimation, temporal compensation and prediction, spatial transformation, quantization, and entropy coding. Still-image compression techniques are a subset of the tools required for video compression. Although current still-image compression standards employ block-transform coding, subband and wavelet methods are now recognized as superior. A new family of biorthogonal 'nearly'-coiflet filters is introduced that possess near-maximal vanishing moments and smoothness (regularity). The design procedure for this family enables time-frequency footprints with for good image-compression performance. Medium-length filters enable superior reproduction of the high-frequency details that are key for video interframe coding. Although often examined independently, quantization, bit allocation and entropy coding may be improved through joint consideration. The performance bounds of Stack-Run coding, a computationally advantageous algorithm for quantization and entropy coding, are examined through the introduction of improved quantizers and the exploitation of two-dimensional spatial redundancy. New space-frequency quantization algorithms enable the relative advantage of joint quantization and entropy coding to be explored. The primary source of redundancy in video is temporal. Compression, the removal of redundancy, is accomplished through the techniques of motion estimation and compensation. This thesis offers two contributions to the temporal processing of video. Although wavelet coding is preferred for lossy image compression, the lack of a block-based intra-block/inter-block switching mechanism prevents efficient coding of video interframes. Space-Frequency block matching is suggested to compensate. Space-Frequency block matching provides a new capability for frequency-adaptive motion-compensated prediction that proves advantageous not only for wavelet interframes, but also for conventional block-transform coding of interframes. Despite success with spatial coding, transform and subband coding are poorly suited to application in the temporal domain. Spatial adaptation that permits both temporal-subband coding and motion-compensated prediction (TSB-MCP) within a single frame spans the gap between the two methods. The hybrid permits coding gain to be sacrificed to increase temporal scalability. A fundamental advantage of the new structure over 'closed-loop' MCP is the simplicity of optimal bit allocation in the 'open-loop'.Ph.D