Adaptive entropy-coded predictive vector quantization of images

Summary form only given, as follows. The authors describe a new approach to image coding based on adaptive entropy-coded 2-D predictive vector quantization (PVQ) ideas. PVQ is a straightforward vector extension of ordinary scaler predictive quantization schemes, such as DPCM, where the vector quantizer (PVQ) is now embedded in the predictive feedback loop. Prediction is then performed on a vector or block basis using previously encoded blocks, with the prediction error blocks subsequently applied, on a block-by-block basis, to the VQ. Although PVQ is not new, previous applications have not attempted to exploit the further compressibility of the VQ output through use of variable-length entropy coding. The authors consider 2-D PVQ of images subject to an entropy constraint and demonstrate the substantial performance improvements over existing approaches. They describe a simple adaptive buffer-instrumented implementation of this 2-D entropy-coded PVQ scheme, which can accommodate the associated variable-length entropy coding while completely eliminating buffer overflow/underflow problems at the expense of only a slight degradation in performance. This scheme, called 2-D PVQ/AECQ, is shown to result in excellent rate-distortion performance and impressive quality reconstructions on real-world images. Indeed, the real-world coding results shown here are rather striking and demonstrate almost imperceptible distortions at rates as low as 0.5 b/pixel.