Pattern-based side match vector quantization for image compression

Abstract

The contiguous pixels are often similar in images. Hence side match vector quantization (SMVQ) can use the contiguous blocks to predict the other blocks. SMVQ has, however, a derailment problem that the error of the current state may result in errors in the next states. To remedy this problem, the pattern-based side match vector quantization (PSMYQ) is used in this paper. Unlike SMVQ, PSMVQ partitions an image into many patterns first. A pattern is a group of n × n blocks. Next, in each pattern, PSMVQ seeds one or more seed blocks, whose corresponding codewords are found by the full search method and employs the side match method to predict the residual blocks. The experimental results show that PSMVQ has three advantages, namely (a) that it is a faster method to encode an image than SMVQ, (b) that it has the ability to reduce the bit rate of an image effectively and (c) that the reconstructed image has a high image quality.