A hybrid data compression scheme using quaternary decomposition and selective multistage vector quantization

The hybrid coding method investigated uses classified vector quantization (VQ) based upon the quaternary decomposition of image blocks into 32×32, 16×16, 8×8, or 4×4 subblocks before using the DCT (discrete cosine transform) to encode the blocks. The coefficients of the DCTs are then zonal-sampled and vector quantized. The codebooks for the VQ are optimized by both the traditional Linde-Buzo-Gray (1980) method and by a simulated annealing (SA) algorithm. In addition a method of selective multistage VQ (SMVQ) is investigated in which only those vectors not matched well enough by a codebook vector are allowed a second stage of VQ. These methods produce excellent results at bit rates of 0.8586 b/pixel. for non-SMVQ methods and 0.994 b/pixel for SMQV methods for 512×512 color images