Advanced Packet-Video Coding Based on Layered VQ and SBC Techniques

Vector quantization (VQ) and subband coding (SBC) are promising candidates for future generations of image coders: VQ is known to outperform scalar-quantization (SQ) techniques, and SBC overcomes the block-structure problem of discrete cosine transform (DCT) coders and has better energy-compaction properties. The paper compares the performances of two SBC-VQ schemes (a quadrature-mirror-filter (QMF)-based SBC with trained-codebook VQ and a parallel-filterbank SBC with lattice VQ) to that of a DCT-SQ scheme as it is used in present image-coding standards. Two-layer versions of these spatial coders were evaluated in a hybrid combination with motion-compensation prediction for interframe data compression. As the SBC scheme with lattice VQ was found to perform the best, this coding scheme was further investigated in combination with more sophisticated interframe-data-compression schemes. The motion-compensated 3-D SBC coder with lattice VQ was found to outperform the techniques used in current standard coders by several dB's in the compression of interlaced CCIR 601 sequences. The performance of this coder is extremely robust in the presence of asynchronous-transfer-mode (ATM) cell losses due to the nonrecursive decoder structure.