A real-time vlsi-oriented architecture for a vq-based color image coding algorithm

There is a lack of real-time image coding architectures for reconstruction of excellent quality color TV images at low bit rates (1.0 bit per pixel). The objective of this research is to develop a novel real-time broadcast quality image coder with an acceptable low bit rate. Vector quantization (VQ) has been demonstrated to be the best approach for obtaining good to excellent quality images at low bit rates. But, high computational requirement of the VQ has hindered its real-time implementation. In this dissertation, different VQ-based image coding algorithms are studied in conjunction with their potential underlying processor architectures for real-time execution. Furthermore, potential modifications to VQ-based coding algorithms to overcome their implementation complexities in light of current VLSI technology and parallel processor architectures are investigated. As a result, a new VQ-based coding algorithm, called mean/quantized residual VQ (MQRVQ), is developed. This algorithm is mapped onto a real-time VLSI architecture. The MQRVQ contributes to the feasibility of the VLSI architecture through use of a simple multiplication-free distortion measure and reduction of the required memory per codevector. The architecture reduces encoding search complexity through partitioning of a large codebook into on-chip memories of a concurrent VLSI chip set. A high-level simulator for the proposed hardware architecture is developed. Simulations are performed using five color TV images. Objective distortion measures and subjective evaluation results demonstrate that a good to excellent quality coded image can be obtained at a fixed bit rate of 1.12 bits per pixel. The architecture is capable of real-time processing of 480 x 768 pixels per frame with a refreshing rate of 30 frames/sec.