Low-power video encoder/decoder using wavelet/TSVQ with conditional replenishment

This paper presents a low-power, low-complexity video encoder/decoder algorithm and architecture for portable video-conferencing applications. We encode each frame using wavelet decomposition followed by tree-structured vector quantization (TSVQ) of crossband vectors, achieving a compression ratio of 16:1 spatially. The temporal correlation between frames is exploited using conditional replenishment. For video commonly used in conferencing applications, conditional replenishment achieves a compression ratio of 4:1 temporally. Thus, the data rate for YUV (4:1:1) compressed video (176/spl times/144/spl times/30 fps) is 150 kbps on average with a PSNR of approximately 33 dB. The proposed scheme has an architectural simplicity, allowing the encoder and decoder to be performed on two separate chips. The power dissipated is estimated to be 7.2 mW for the encoder and 1.0 mW for the decoder in a 0.8 micron CMOS technology at 1.5 V.