Multiple motion-compensated and wavelet-based video compression.

摘要

This research deals with the problem of transmitting a video image sequence over a channel, whose bandwidth is limited. A compression scheme that is comprised of motion compensation, transform, quantization and entropy coding is adopted.; Our contribution to this research is focused on: (1) A fast and flexible algorithm for motion compensation; (2) Application of wavelet transform to compress the residue images; (3) A fast quantization strategy for wavelet transformed residue images.; In this dissertation, the classical motion estimation methods are reviewed. The new approaches based on a greedy strategy are presented. Experimental results are used to demonstrate better motion compensation as well as reduced computation time. Results suggest that this approach can handle multiple motion at different 2-D velocities.; More compression can be achieved by transforming the prediction errors and then quantizing the transform coefficients. Wavelet transform can achieve a better compression rate over DCT due to its adaptability to the narrow width and narrow band signals.; The quantization of the transform coefficients is another design problem. While the operation of quantization is simple, the design of quantizers and boundary values can be difficult and relies on the model of the data source. In our case, the data source is the wavelet coefficients on the residue image after motion compensation. It is nonuniform and highly uncorrelated. Max-Lloyd quantization technique, which has been proved to be optimum for a nonuniform source with known statistics, is discussed. The modified Max-Lloyd quantization algorithm, which deals with discrete data and improves the speed of convergence, is presented. Results are also shown to demonstrate the effect of the modified quantization scheme.