Fully scalable subband/wavelet video coding system

In this thesis, a fully scalable video coding system is presented, which utilizes a motion compensated 3-D subband/wavelet transform to realize temporal, quality and resolution scalability.; A directional IBLOCK is proposed in motion compensated temporal filtering (MCTF) framework to process the areas with difficulty in motion estimation or without temporal correspondence, which significantly benefits medium and low bitrate video. The motion path for a pixel in a block is correct only with some probability in hierachical variable size block matching (HVSBM). We propose an overlapped block motion compensation (OBMC) with modified bilinear probability windows for the various block sizes, a shrinking scheme to reflect motion discontinuity, and an optional further iterative optimization procedure, all in the lifting filter implementation. With OBMC, the video is almost free of block artifacts giving a significant objective and subjective improvement. We also investigate the MCTF pyramid structure in temporal decomposition and extend HVSBM on Y, U and V components simultaneously, called color HVSBM. We combine color HVSBM, directional IBLOCK, and OBMC, and find almost additive improvements from the three additions.; The algorithm of frequency roll-off via energy matching is proposed to reduce the spatial aliasing in the extracted low resolution video from a scalable subband/wavelet video coder. Frequency roll-off offers significant visual improvement for low resolution video but introduces a little or no loss for full resolution video.; We propose a new scalable motion vector coder based on context based adaptive binary arithmetic coding (CABAC). A selective layered structure and alphabet general partition (AGP) of motion vector symbols are employed for SNR and resolution scalability of the motion vector bitstream. With the two methods and the temporal scalability already existing, we have a basic format of temporal, SNR, and resolution scalability for the motion vector bitstream, which improves significantly both visual and objective results for low bitrate and low resolution with only slight PSNR loss and unnoticeable visual loss at high bitrate.; We extended our thesis work to MCTF framework with adaptive switching between LeGall 5/3 filers and Haar filters.