An adaptive vector quantization based on neural network

Abstract

Some vector quantization algorithm are first surveyed. Then, an adaptive vector quantization method for image coding based on a neural network is proposed. This method first partitions the image into a subimage and transforms them with the DCT, and then classifies and encodes them in the transformed domain using frequency sensitive competitive learning (FSCL). The experimental results show that this VQ method has no local region distortion and a high compression ratio.