Abstract：A new method to perform classified vector quantization on wavelet images using directional tree structure vector combination is proposed. It makes vector combination by employing the direction character of subband coefficients and the correlation between and within subbands. It has two steps of classification with the vector’s energy and activity and thus reduces the inner dispersion of the classified vectors. The classified information needs few bits to carry. The weighted mean square error rule using human visual characteristics is used to improve quantization gain. Simulation results show its coding efficiency for wavelet images.
2005, Vol. 27 
