Adaptive subspace image vector quantization and its relationship with transform coding using VQ

Novel complexity reduction methods for spatial domain image vector quantization based on subspace distortion is proposed. Substantial reductions in computation and memory requirement are obtained while maintaining good image quality. Experimental results show that the fixed-basis subspace distortion method can achieve as much as four times improvement in both computation speed and memory requirement with an image quality degradation of not more than 0.4 dB in peak signal to noise ratio for many real images, The adaptive-basis subspace distortion method can achieve almost 16 times complexity reduction for the case of binary tree-searched VQ. It has further been shown that the proposed subspace VQ method is always better than a corresponding transform domain VQ having the same complexity. The proposed methods are general and can be applied in combination with many other image VQ techniques to achieve further improvements