Low bit-rate video coding using space-frequency adaptive wavelet transform

In the still picture coding, we proposed a Space-Frequency Adaptive Coding (SFAC) approach, which is a zerotree based coder with a new optimization algorithm in the rate-distortion sense. In this paper, we extend the SFAC to video coding, in which the zerotree quantization is applied for both intraframe and inter-frame in order to trace energy compaction in motion residual image. To remove the temporal correlation efficiently, motion compensation is performed in image domain rather than in wavelet domain. The experimental results with the standard video image sequences, such as Football, Salesman and Claire, demonstrate that higher performances both for intraframe and interframe are achieved with zero-tree optimization. Compared to the conventional techniques, MPEG 2 and other wavelet video coders, 2–3 dB improvement in Football(SIF 760 kbps, 30 fps), 2–5 dB improvement in Salesman(QCIF 30 kbps, 10 fps) and 3–6 dB improvement in Claire(QCIF 30 kbps, 10 fps) can be obtained. The improvement is remarkable in lower bit-rate coding and the improved subjective quality can be observed in decoded image sequences.