Video Coding Using Global/local Motion Compensation, Classified Subband Coding, Uniform Threshold Quantization and Arithmetic Coding.

This dissertation is devoted to the design of a generic video coding algorithm which has the flexibility to be applied to a variety of applications. It can compress video signals with high performance at almost any fixed output rate or distortion. The algorithm consists of four components, each designed to target at a specific characteristic of video signals. We introduce global/local motion compensation to exploit temporal correlation. Unlike conventional local motion estimation approaches, it has an additional stage which uses a three-parameter model to account for global motion due to camera zooms and pans. Parameter estimation procedures based on either difference frame or displacement vectors are given. Both demonstrate high accuracy. The latter has a faster convergence rate. The effectiveness of our approach is demonstrated by an improved prediction quality with reduced motion information, and the existence of simple implementation requiring only a nominal increase in complexity. We combine subband coding with spatial classification to exploit spatial correlation. A theoretical analysis of variance-based classification is developed, which provides optimal threshold setting. Simulation results show that the classified subband approach offers a slight improvement over conventional subband approaches at high rates. To remove perceptual irrelevancy in the signal, we use an entropy-constrained scalar quantizer known as uniform threshold quantizer. It has a rate-distortion performance comparable to the optimal scalar quantizer, and lies no more than 0.3 bits/sample above the rate-distortion bound for many sources. Its structural regularity simplifies design procedure and provides the flexibility we require. Arithmetic coding is used to exploit the non-uniformity of source distributions. Growing precision and carry propagation are solved by integer approximation and our new optimal code interval adjustment, respectively. Our code uses fixed precision arithmetic and buffer, and has near optimal compression performance over a wide range of sources. Simulations show that the video compression algorithm consisting of the four components above meets our performance and flexibility objectives. In addition, the techniques we develop for each component can be applied to many other compression algorithms.