Image compression using discrete wavelet transform and classified vector quantization

Abstract

Two major contributions of this thesis are analysis of discrete wavelet transform(DWT) basis for very low bit rate image coding and introduction of a new image coding scheme that is based on DWT and classified vector quantization(CVQ). The proposed system achieved high compression ratio with good image quality as well as very simple architecture. We conclude that regularity and linear phase properties are the most important conditions that DWT basis should have. New coding scheme proposed in this thesis shows better performance than most coding scheme already published. The proposed coding scheme divides the wavelet coefficients into DC and AC coefficients, where DC coefficients are coded independently from AC coefficients. The AC coefficients are rearranged to be vector quantized, and the rearranged AC coefficient vectors are classified by exploiting the characteristics of DC coefficients corresponding to the AC coefficients vector. Progressive transmission of wavelet coefficients, i.e., transmitting DC coefficients first and AC coefficients next, makes decoder extract the classification information exactly from the already transmitted DC coefficients. Therefore it is not necessary to assign extra bits for class identification and overall coding performance can be improved. Method of classification and codebook generation was very important in this scheme. The method reduces the bit needed to code images significantly. The advantage of the scheme in this thesis is the simplicity of the architecture of hardware. It does not use any adaptive techniques that make decoder complex. Simple hardware strategy was pursued in all parts of scheme. In order to verify the proposed coding scheme has high performance, several comparison studies were given through computer simulations.