Encoding of colour still pictures by wavelet transform and vector quantization

This paper presents research results on colour-picture compression by wavelet transform and three-level quad-tree coding. First, a still picture based on an RGB colour model is converted to YIQ and YUV colour components. In the encoding process each colour component is encoded individually. A colour component is decomposed to (3 × 3 + 1) subband images using the 2-D discrete wavelet transform three times. Next, the L3LL subband image having the lowest-frequency component is encoded by DPCM. The remaining nine subband images are encoded by a scheme based on zero-value quantization. The zero-tree representation of the coefficients belonging to an area of three different resolution levels is then implemented, and the zero-tree run length is MH-encoded. The coefficients are optimized by five-dimensional vector quantization based on the LBG algorithm, which is used to encode the remaining coefficients not included in the zero tree, and only optimized coefficients are thus encoded. In other words, improvement of image quality for reconstructed images is established because additional non-zero-tree information has been encoded. These results, including the comparison of each filter, show the effectiveness of colour-picture compression by wavelet transform and three-level quad-tree coding.