Predictive tree encoding of still images with vector quantization

The superior performance of tree encoding and vector quantization (vq) over scalar quantization (sq) is well known. However, large constraint length (in tree encoding) and large vector length (invq), which are required for a close-to-optimum performance, are limited by the exponentially growing complexity of the source encoder. The intention of our work has been to combine tree encoding andvq, each of moderate complexity, to preserve the advantages of both block and convolutional codes. This principle was applied to the prediction error signals of images as produced by non-adaptive and adaptive linear predictors.