Raster Search Resolution Analysis for Fast Motion Estimation in HD Video Compression
Pavel Arnaudov, Tokunbo Ogunfunmi · 2018
HD Video Motion estimation is the main part of both video compression standards - HEVC and H.264 in terms of power and processing time. Fast Motion Estimation Algorithms use techniques to reduce the computation with a minimal quality loss. Often the Search Process is performed in several phases through only a subset of the possible options to reduce time, power and cost. In the case where quality is critical, the first pass can be a more detailed search, the step of which has quadratic impact on the computation and from there on power. This paper presents an analysis of the optimal steps for HD video, which allow reduction in computation in the range of 16-89%, just before the quality starts to drop significantly. Full Search quality is the ultimate goal, but its computation is orders of magnitude higher than Fast Motion Estimation methods. The search step is applied in both X and Y directions of an HD frame search, so its impact on computation and power is quadratic in time and complexity. The pervasiveness of battery operated devices like cell phones with HD cameras, which need to process, store and transmit HD video, emphasizes the importance of this analysis. Our analysis shows that the optimal step before the start of a noticeable quality loss is in the range of 6-8 pixels.