Predictive-transform residual vector quantization

This work investigates a new compression scheme that incorporates rate-constrained residual vector quantization (RVQ) of the transform coefficient error vectors arising in a minimum mean squared error (MMSE) predictive-transform (PT) methodology. The proposed technique that we call predictive-transform residual vector quantization (PTRVQ) exploits the correlation between neighboring blocks by predicting the transform coefficients of each block using information from previously encoded blocks. PTRVQ, therefore, exploits both the intra- and inter-block correlation as well as employs dimensionality reduction in the transform domain in order to generate a highly compact code. Experimental results obtained with several standard monochrome images and compression ratios as high as 135:1 demonstrate quite noticeable improvements of the reconstructed images in both perceptual quality and peak signal to noise ratio (PSNR). The PSNR improvements are as high as 1.4 dB over a wavelet based coder.