Video compression using vector quantization for multimedia applications

In this dissertation, three novel video-compression algorithms are presented for real-time, very-low-bit-rate applications with operating bit-rates ranging from 5 to 16 kbps. The algorithms are based on different forms of vector quantization (VQ) techniques: adaptive residual vector quantization (RVQ), quadtree-based vector quantization, and cache-based vector quantization (a form of finite-state vector quantization). All the three algorithms outperform H.263, the current very-low-bit-rate video coding standard, in the rate-distortion sense; at the same time, they avoid blocking artifacts, which is one of main drawbacks of H.263.; The technical requirements of video coding algorithms, such as real-time constraint, scalability, complexity, and rate-distortion performance, differ from one another according to the target applications, such as video multicasting, video interactive systems, and video conferencing. Accordingly, the choice of an appropriate video coding algorithm depends on the application environment in which the selected video coding algorithm will operate. The three video coding algorithms have different target applications according to the unique advantages of each. The following key technical factors are considered when developing the video coding algorithms: (1) Capability for high compression while suppressing annoying artifacts such as blocking and mosquito artifacts. (2) Simplicity of the encoding and decoding system. (3) Scalability in terms of bit rates and picture quality. (4) Compatibility with existing video coding standards (if necessary).; The adaptive RVQ video codec satisfies all the items listed above. The RVQ codec is a multistage residual vector quantizer with transform vector quantizers in the initial stages. Its powerful bit allocation scheme, based on variable-rate RVQ, provides better rate-distortion performance over H.263 while suppressing annoying blocking artifacts. The quadtree-based video codec also provides better rate-distortion performance over the H.263. An optimal quadtree decomposition of the signal for a fixed vector quantizer is also presented to achieve optimal quantization at a given bit rate. The system complexity of the quadtree-based codec is also much simpler than those of the H.263. The picture quality of the cache-based VQ video codec is a significant improvement over previous codecs in terms of annoying distortions (blocking artifacts and mosquito noises), and is comparable to that of recently developed wavelet-based video codecs. The simplicity of the encoder and decoder of the cache-based VQ codec makes it more suitable than wavelet-based coding for real-time, very-low-bit-rate video applications.