Multi-level Parallel H.264 Encoder For Multi-core Architectures

With the rapid development of multimedia technology, the transmission and sharing of the high-quality video is demanding urgently. H.264, widely used by high-definition video, has the advantages of high-quality and low bit rate, but the complexity of encoding increases greatly. How to speed up H.264 encoding without lowering the video quality and increasing the compression ratio is an important hot research problem. With the continuous development of multi-core processors, the speed of the serial H.264 encoding programmer has been unable to meet this real-time transmission and large-scale sharing requirements. The development of the multi-level parallel H.264 encoder on multi-core processors environment has great significance to the current H.264 video encoding.Based on the existed research developments, the multi-level parallel H.264 encoder which combines the frame level, the slice level and the data level is proposed with the necessity and feasibility analysis on multi-level parallel H.264 encoding. Using the open-source H.264 encoder project X264, an adaptive B-frame frame parallel method is proposed with respect to the complex relationship among the reference frames. Base on the study of the SPEs on the heterogeneous multi-core platform, a data level parallel method, which splits a 16x16 macroblock into several parts fitting parallel on SPEs, is designed. According to the characteristics of homogeneous and heterogeneous multi-core platforms, the multi-level parallel H.264 encoder is implemented on these two platforms by designing the three data buffer queues and creating multi-level threads.Experimental results show that along with the upgrading of the encoding speedup to a large extent by the multi-level parallel H.264 encoder, the bitrates do not increase obvious clear. Beside this, encoding video quality in line with the characteristics of high-quality, accesses to the superior performance of the encoding. However, there still has two things need to be improving, one is the speedup on the heterogeneous multi-core platform, the other one is the fine allocation of resources on the multi-core. The future work will focus on these two directions.