Side-Match Vector Quantizers Using Neural Network Based Variance Predictor for Image Coding

Abstract

Side-match vector quantizer reduces bit-rates in image coding through exploiting the correlations of neighboring vectors. This paper presents a new side-match vector quantization method for image coding using a neural network-based variance predictor. In this method, the master codebook used for generating the state codebooks in side-match vector quantization is sorted according to the variances of the codewords. Unlike the regular side-match vector quantization which side-matches all of the codewords in the master codebook to select codewords to construct the state codebooks, the proposed method side-matches subsets of the codewords in the master codebook, selected based on the variance of the current vector being encoded. The variance of the current vector is predicted by a feed-forward three-layered neural network. Experimental results demonstrate that in terms of PSNR (Peak Signal-to-Noise Ratio) of the reconstructed images, the proposed method significantly outperforms the regular side-match vector quantizer, especially at lower coding bit-rates.