Research And Optimization On H.264/AVC Real-Time Video Encoder

H.264/AVC is an emerging video standard by Joint Video Team. It is aimed at high video quality at low bit-rates. But the computation complexity of the codec is very high, especially the performance of the encoder. This limits its application area. Due to this cause, many researchers tried their best to study the H.264 encoder to speed up it. However, they mainly concentrated on algorithm or instruction level optimization, such as the motion estimation algorithm optimization, mode decision algorithm optimization, and SIMD instruction optimization of the time-consuming models, etc.Firstly, this paper introduces the basic concept, primary property and present situation of research of the H.264, and the basic knowledge of digital video encoding. Then we talk about the motion estimation and mode decision algorithm, and analysis the motion estimation algorithm of H.264 reference mode software JM. Finally we focus on the optimization tactics of the JM Encoder and the implementation of the H.264 real-time encoder. The main contribution of this paper is that we propose a solution of H.264 real-time encoding based on symmetric multiprocessor cluster. This solution realizes two levels parallel encoding: slice level and macroblock level. The system is achieved by distributed of shared memory architecture. We implement it by hybrid programming of MPI and OpenMP. The reference H.264 encoder is JM10.1 provided by JVT. As the experiment shows, the optimization of architecture speeds the H.264 encoder's performance highly and can apply this kind of encoder to real-time encoding environment.