Vector quantization image coding based on biorthogonal wavelet transform and improved SOFM

Abstract

This paper studies the statistical properties and distributed properties of the coefficients after the image is decomposed at different scales by using the wavelet transform. The different quantization and coding scheme for each subimage are carried out in accordance with its statistical properties and distributed properties of the coefficients. The wavelet coefficients in low frequency subimages are compressed by using Differential Pulse Code Modulation (DPCM). The wavelet coefficients in high frequency subimages are compressed and vector quantized by using Kohonen neural network on Self-Organizing Feature Mapping (SOFM) algorithm. In addition, an improved SOFM algorithm is used in vector quantization in order to shorten the encoding and decoding time. Using these compression techniques, we can obtain rather satisfactory compression ratio as well as shorten the encoding and decoding time while achieving superior reconstructed images.