Visual Media Accelerating On Multi-core Architecture

Real-time processing of visual media such as graphics, image and video is one ofthe crucial research domains in multimedia and parallel processing. Conventionally,visual media research focuses on how to improve the e?ciency of visual media algo-rithms. With the improvement of e?ciency of visual media algorithms, the computingcomplexity increases significantly, and the low processing speed renders most visualmedia algorithms outside of the realm of practicality. The present thesis focused on theresearch of parallel optimization for visual media algorithm. The thesis summarizedthe development of multi-core processor and the parallelization of media processing al-gorithm. Based on the parallelization of ray tracing and multi-view video coding overmulti-core platform, the thesis proposed a visual media dynamic scheduling frameworkbased on multi-core architecture and an improved multi-view video coding algorithmwith high parallelism. The main contributions of the thesis are the following:1. The thesis proposed a parallel ray tracing over hybrid multi-core processors (IBMCell/B.E.). The proposed parallel system chose data parallel control mode andKD-tree as spatial index structure. The thesis analyzed the parallelism and com-puting complexity of the parallel system which obtained scalable speedup.2. The thesis proposed a static scheduling performance estimation method formulti-view video coding algorithm, and defined evaluation parameters for par-allelism, e.g. concurrent degree. Subsequently, the thesis proposed and imple-mented an adaptive multi-view video coding acceleration algorithm over hybridmulti-core processors (IBM Cell/B.E.), and a frame-level parallel multi-viewvideo coding acceleration algorithm over symmetric multi-core processors (Intelmulti-core processor). Both of them obtained scalable speedup.3. The thesis proposed a dynamic scheduling performance evaluation frameworkfor visual media processing. Based on directed acyclic graph theory, the frame-work was able to evaluate the parallel processing performance of visual media over multi-core processor without complete implementation of parallel applica-tions. Therefore, the framework could simplify the design progress. In the the-sis, we made parallel acceleration system of multi-view video coding over Intelmulti-core processor as an example, and proved the validity and simplicity of theframework.4. The thesis proposed a multi-view video coding algorithm with high parallelism.Experiences from parallelization were applied to multi-view video coding algo-rithm design which promoted the parallelism of multi-view video coding throughimproving prediction structure.