Algorithm and VLSI Design of a Feature-Based Classi ed Vector Quantizer for Image Coding

In this paper, a feature-based classi ed vector quantization (FCVQ) algorithm and VLSI implementation of the classi er are presented. The FCVQ technique exploits the characteristics of discrete cosine transform (DCT) and the concepts of block truncation coding (BTC) to simplify the classi cation and preserve the edge information e ciently. During the classi cation, an input block is classi ed as an uniform block or a block with a speci ed orientation of edges at rst; di erent partition schemes are used for encoding di erent types of blocks. There are only simple additions required for the classi cation. Simulation results show that the approach alleviates edge degradation in primary VQ with bit-rates as low as 0.35 bits per pixel (bpp). The proposed classi er has been implemented by 0.8m CMOS technology on a 6.3 mm 4:7mm die with a clock rate of 35 MHz.