Joint temporal-spatial bit allocation for video coding with dependency

The joint temporal-spatial bit allocation problem with consideration of dependency arising from motion compensated prediction as well as frame interpolation is investigated in this research. After the problem formulation, several heuristic methods are proposed to provide near-optimal and suboptimal solutions to this problem. First, the selective iteration algorithm (SIA) based on the monotonic property of the rate-distortion (R-D) curve is proposed to achieve the near-optimal solution. Then, the greedy iteration algorithm (GIA) is presented as a suboptimal heuristic to reach an R-D performance close to that obtained using SIA, while at a much lower complexity. Furthermore, by adaptively grouping frames based on the mean of the absolute difference differentials and applying greedy pruning to groups of frames, the suboptimal solution is significantly expedited so that it can be applied in real-time applications. Frames to be skipped in the coding process and quantization parameters (QPs) exploited in coded frames are adaptively and jointly determined to reach a proper tradeoff between temporal and spatial qualities. Experimental results show that the proposed methods can enhance the overall quality of compressed video at various bit rates in comparison with H.263+/TMN8 using fixed frame rates and QPs, as well as the adaptive quantization solution proposed in.