A Wyner-Ziv video codec with horizontal and vertical sampling

In Wyner-Ziv (WZ) video coding, decoder side information (SI), among other components, takes a key role in decoding and reconstruction. In most existing WZ video coding schemes, SI is generated using motion compensated temporal interpolation (MCTI), which requires very frequent insertion of intra (key) frames in encoded bit stream or a very small group of pictures (GOP) size. This makes such methods far from practical, and enlarging GOP size is helpless to overcome this drawback as this will reduce the temporal correlation between intra frames and hence degrades the performance of MCTI. To overcome this drawback, we propose, in this paper, a WZ video codec with horizontal and vertical (HV) sampling. In the codec, a frame, before coding, is divided into four equal-sized sub-frames using an HV sampling process. Only one of the sub-frames is intra coded, while the remaining ones are WZ coded. This way, spatial correlation in an intra frame is converted to temporal correlation, while temporal correlation between adjacent frames is partly converted to spatial correlation, and SI at the decoder can always be estimated through motion estimation using the intra coded sub-frames. The tight constraint on GOP size of other WZ video coding schemes is released in the proposed WZ video codec, and hence leads to more practical WZ video coding solution.