Vector subband video coding

As the exchange of information in our society grows, efficient video compression is rapidly becoming an integral part of most video systems. Scalar subband coding (SSC) uses a conventional scalar filter bank (SFB) in order to break an image into more localized units called subbands for more efficient coding. A process called Vector Subband Coding (VSC) maximizes the performance of vector quantization (VQ) by "vectorizing" SSC. The VSC process results in a high quality image at a low bit rate by preserving intra-vector correlation while reducing inter-vector correlation. Memoryless VQ is not optimal in the sense that it does not use past information in making a decision about the coding of a present input vector. Finite-State Vector Quantization (FSVQ) can be implemented to incorporate memory into the VQ system. Using a finite number of states, the FSVQ system utilizes information on previously reproduced vectors and past states to determine how to process the current input vectors. One contribution of this study is to use the FSVQ concept for interframe video coding. Two different techniques have been developed. The first one is equivalent to a first order prediction system without using motion compensation. The second technique is equivalent to a second order prediction system using motion compensation. In addition to compression, a coding system with progressive transmission is desired for many emerging applications, such as Internet and multimedia. The flexibility in this type of transmission allows for precise rate control and a progressive increase in video quality based on the capacity of the channel. A contribution of this study in this area is to consider the popular discrete-cosine transform (DCT) to be a filter bank and understand the relationship between video quality, the Q value, and the number of frequency hands. Another contribution of this study is an efficient implementation scheme for filter banks.