Optimal signal processing for vector quantization based image and video coding

In this dissertation, recent research results on optimal signal processing for vector quantization (VQ) based image and video coding with complexity constraint are presented.;Vector quantization always outperforms scalar quantization. Signal processing before vector quantization can greatly reduce VQ complexity (both dimensionality complexity and bit-allocation complexity) and improve the overall rate-distortion performance. Based on the asymptotic quantization theory, performance of vector quantization especially for low bit rate cases is studied. A performance formula is derived to estimate the rate-distortion performance of low bit rate VQ.;A general framework of joint performance analysis of signal processing combined with vector quantization is established. Within this framework, the rate-distortion performance and VQ complexity of different signal processing schemes for both 1-D and 2-D signal sources are analyzed for both high resolution and low bit rate quantization. A modified BFOS optimal bit allocation is presented for low bit rate coding. Performance of several existing signal processing schemes combined with VQ are evaluated and compared under the complexity constraint.;A set of design rules for optimal signal processing schemes are present and a new mirror-sampling based vector transform is proposed. It is proved that the proposed vector transform achieves the optimal performance in high rate case. The properties of hierarchical decomposition using the proposed mirror-sampling based signal processing are studied. Applications of these theoretical results to image and video coding are also presented to support our theoretical analysis.