High Definition Real Time Scalable Video Coding Based On Tilera-Gx Processor

With the development of video applications, the H.264/AVC video coding standard has a wide range of applications by the virtue of its high compression ratio. As the appendix of H.264, the Scalable Video Coding(SVC) has great prospects because it can provide different bit streams for different calculating platforms and different display devices to meet different client requirements. However SVC is extremely difficult to be applied in high definition real time fields due to its high complexity. Currently, the rapid development of multi-core processor provides sufficient support for high complexity computing. Based on the analysis SVC application requirements and the Tilera multi-core processors’features, a multi-core parallel SVC encoding scheme for HD video is developed.This thesis focuses on the SVC encoder software development and multi-core implementations. Based on the analysis of algorithms and computing platforms, a Tilera-Gx-oriented SVC parallel coding algorithm for HD video is proposed. Firstly, among layers, considering the temporal hierarchy of SVC, this thesis implements temporal layer aligned parallel coding in spatial level by analyzing the course of different performance of T3layer and other layers. Secondly within a layer, considering the actual encoding time of slices in one frame, PID control method is used to dynamically determine the MB number of a slice, and reduce the impact of diversity of image content to encoding performance. As a result, the dynamic equilibrium of different cores achieved. Thirdly, since the LPF module dependence, a parallel loop filter method is proposed. At last, in order to take good advantage of cores, a dynamic core allocation scheme is proposed. Experimental results show that the proposed solution achieves30fps encoding, and speeds up the encoding process by more than19times. It has implemented the real time encoding for720p high definition sequences, which is of high usability.Finally, this thesis is concluded and gives the prospects of the higher level optimization of SVC for the future.