RAVIOLI: A parallel video processing library with auto resolution adjustability

Video processing applications are now in demand on a great variety of platforms such as mobile devices or high performance servers. On the other hand, a great variety of performance is required for video processing applications such as high throughput, good accuracy, long battery life, and so on. Therefore, programmers today should study hard about various platforms and techniques under the pressure of necessity. This must make the burden heavy for them. This paper proposes a parallel video processing library RaVioli. RaVioli achieves self-optimizations for multi-core processors and self-adjustment of resolutions. RaVioli conceals two resolutions, frame rate and number of pixels, from users and provides dynamic and transparent resolution adjustability based on user-preferred priority parameters. This makes pseudo real-time video processing feasible for any platform by adjusting resolutions according to situations. Generally, video processing has some parallelism in its algorithm. For example, pixels in a frame have data parallelism, and many video processing algorithms can be divided into some processing stages which can be pipelined. Concealing resolutions makes implicit parallelism more obvious. Hence, RaVioli can parallelize programs semi-automatically.