Video Compression using Vector Quantization

This report presents some results and findings of our work on very-lowbit-rate video compression systems using vector quantization (VQ). We have identified multiscale segmentation and variable-rate coding as two important concepts whose effective use can lead to superior compression performance. Two VQ algorithms that attempt to use these two aspects are presented: one based on residual vector quantization and the other on quadtree vector quantization. Residual vector quantization is a successive approximation quantizer technique and is ideal for variable-rate coding. Quadtree vector quantization is inherently a multiscale coding method. The report presents the general theoretical formulation of these algorithms, as well as quantitative performance of sample implementations.