Subband image compression using wavelet transform and vector quantization

The Discrete Wavelet Transform (DWT) has many useful properties when applied to image compression. The multiresolutional decomposition is of complexity O(n) and conserves the geometric image structure within each subband. A tree-structured coding scheme can efficiently exploit the inherent correlation in the subband representation. An algorithm is introduced which incorporates a fast tree-structured quantization scheme and partial search vector quantization. The algorithm described in this paper is a novel quantization thresholding scheme which uses the DWT to decompose an image into octave wide frequency bands, then quantizes the coefficients using a “look ahead” measurement of the image based on the low frequency sub-image inherent in the DWT. This algorithm then uses vector quantization to code the thresholded coefficients of the decomposed image. A partial search vector quantization algorithm is used to increase the speed of the quantization by using a sorted table of the energy content of the code vector. Each subband has an associated codebook which is generated using the Pairwise Nearest Neighbor (PNN) algorithm to produce an initial codebook and then uses the generalized Lloyd (GL) algorithm to arrive at a final codebook