The Technology Of Video Coding Based On Nonseparable Wavelet Transform

Wavelet applying to image compression has been researched widely because of its characteristic of local time/frequency. There are many excellent encoding shemes such as EZW, SPIHT, SPECK, EBCOT. JPEG2000 has accepted EBCOT as encoding part, realized gradually transferring, ROI (region of interest), etc, sides powerful compressing.Despite of its widely use in still image compression, the research on applying it to video coding has lagged behind a little. However, many researchers engage in this research region. MPEG-4 has accepted wavelet transform as its objected oriented encoding scheme. So, it can be seen that with the appearance of more excellent encoding scheme and the advancing of hardware, wavelet will definitely be used in video compression.There are three branches in the wavelet based video encoding now: OBMC-DWT, DWT-MC and 3D-DWT. Each of them has its own superiority. The former two is researched in this paper.As a different characteristic from traditional separable wavelet, the nonseparable wavelet transform doesn't distinguish direction, and it can predict a current pixel using its adjacent pixels in all directions with lifting scheme. So it has superiority compared to separable wavelet in image coding. There have been some researches on applying it to image compression.This paper applies the nonseparable lifting wavelet transform based on APDICSF and the 2-branch SPIHT to video compression. First, the OBMC-DWT video coder applying appropriate filters for I and P frames is constructed and the PSNR of the resulted image is compared with the separable video coder by experiments. Then, the DWT-MC video coder is also constructed. Utilizing the directional characteristic of the nonseparable wavelet, two more compensation schemes in wavelet domain are presented. The search ranges in lower and higher bands are given by experiments. The PSNR is also compared with the separable wavelet video coder.