Variable block-sized vector quantization of grayscale images with unconstrained tiling

The design and performance evaluation of a variable block-sized vector quantizer with unconstrained tiling (UTVQ) is discussed. The coder segments the input image into an unconstrained tiling of variously-sized, nonoverlapping rectangular regions, each of which is then coded with a vector quantizer. UTVQ is compared with the more constrained quadtree vector quantizer (QTVQ) approach. A transmission scheme which reduces the overhead required to convey UTVQ segmentation geometry is developed. Several UTVQ segmentation strategies are explored, and mean-residual codebooks with appropriate side-stream compression techniques are developed. The projected rate and rate-distortion figures indicate that the UTVQ system outperforms the QTVQ system under typical conditions.<<ETX>>