Subband coding of images using transform vector quantization

A subband coding scheme (SBC-TVQ) for image coding in which vector quantization (VQ) is implemented in the discrete cosine transform (DCT) domain is presented. SBC-TVQ exploits the correlation in large image blocks by taking advantage of the compaction of the energy along certain directions in the transform domain of the high-frequency subbands. In this method, the spectrum of the input image is decomposed nonuniformly into seven bands using a quadrature mirror filter (QMF) bank. For each of the three high-frequency subbands, the transformed blocks are classified and adaptively partitioned into smaller subvectors with different dimensions. The obtained subvectors are quantized using VQ. The low-frequency subband is coded with higher precision using the classical adaptive DCT coder. Reasonably good reconstructed images are obtained at a rate around 0.4 bit per pixel.