An optimized hybrid Vector Quantization for efficient source encoding

Motivated by the efficiency of the well-known zero-tree scalar coding of wavelet transformed images, several attempts have been made recently to adopt similar methodologies in Vector Quantization (VQ). This latter approach, however, fails to retain fine details with a reasonable codebook size. We present a novel hybrid multiscale VQ (HMVQ) coder to preserve image details by creating a universal codebook of modest size with a combination of VQ and residual scalar coding of selective wavelet coefficients. Such codebooks employ an efficient clustering technique to ensure minimal rate distortion even at very low bit rates and could be used for a compact and cost-effective system design when used in an energy optimized wavelet sub band domain.