Auto-Parallelization for a Video Processing Library with Content-Aware Resolution Management

The performance of general purpose computers is increasing rapidly, and now they are capable of running video processing applications. However, on general purpose operating systems, real-time video processing is still difficult because there is no guarantee that enough CPU resources can surely be provided. A pseudo real-time video processing library Ra Violi has been proposed for solving this issue. Ra Violi conceals two resolutions, frame rate and number of pixels, from programmers and provides a dynamic and transparent resolution adjustability. Namely, Ra Violi regulates the processing load by automatically modifying two resolutions according to CPU utilization. Besides, Ra Violi can divide a whole video frame into several sub-frames, and process each sub-frame in different precisions considering its importance. Using Ra Violi, pseudo real-time video processing can be achieved easily, but output precision may be extremely roughened for reducing processing load even if each sub-frame is processed with the suitable precision. To solve this problem, we propose an auto-parallelization mechanism on Ra Violi. We aim to improve the processing precision with the mechanism.