Subband/VQ Coding of Color Images Using a Separable Diamond Decomposition

Abstract The combination of subband coding and vector quantization can provide a powerful and flexible method for compressing color images, since this type of system can easily incorporate properties of the human visual system (HVS). Most of the previous work in image compression using subband coding has filtered the input image so that the frequency domain is partitioned into rectangular regions. While this rectangular decomposition allows one to use separable filters resulting in an easy implementation, it does not take into account the higher sensitivity of the human eye to horizontal and vertical edges than to diagonal ones. In this work, we present a new method for coding high resolution color images using a separable diamond subband coder and vector quantization. By first diagonally interpolating the input image and then filtering with one-dimensional filters, a decomposition into subbands that more closely match the orientation of the human visual system is obtained. Simulation results are given for several vector quantizer cases. Performance evaluations show that for compression ratios in the range of 10:1 to 20:1, a five band system with diamond-shaped subbands yields results that are comparable to those of a seven band rectangular system.