Vector quantization for image compression using circular structured self-organization feature map

Abstract

We propose a stable and robust vector quantization coding scheme for image compression known as circular self organization feature map (CSOM) by introducing circular structure to a basic codebook. This structure enables the self organization feature map (SOM) method to converge faster, and to learn input vectors more efficiently. The results suggest that CSOM gains approximately 30% speedup in computation time and 0.3 dB in the PSNR compared to the conventional SOM algorithm. In addition, robustness for initial state of a codebook is achieved by CSOM.