Improvement Of H.264 Coding And Its Implementation On Web Server Of Loongson Platform

H.264/AVC is widely used in the communications and Internet industries due to its efficient compression ratio and strong network adaptability.However,its excellent performance is achieved through the introduction of many new coding techniques,which improve the coding performance of H.264 on the one hand,but on the other hand,it brings huge computational complexity.For example,it is difficult for H.264 to exploit full advantages on some embedded platforms with limited resources,which is also a obvious defects.This paper analysed the coding complexity brought by several key technologies in H.264,and select intra prediction and inter prediction to improve.In order to solve the over long encoding time problem in the process of intra prediction mode selection,an improved fast intra prediction mode classification algorithm was proposed.Through the analysis for the characteristics of high frequency components in frequency domain,it can be known that the fast separation of intra 4×4 and intra 16×16 model can be realized by using training threshold.At the same time,considering the problem of over many intra 4×4prediction models,the improved Kirsch operator can be used to characterize the texture direction of the image,so that the 4×4 prediction model was reduced.The experimental results under the JM model show that the improved intra prediction model fast classification algorithm reduced the coding time by 67.81%.In addition,in order to solve the high time consumption of the Umhexagon S algorithm in the optimal motion vector estimation.Firstly,after obtaining the best point in the searching period,shifting into three-step iterative motion scene classification of the EDR model directly.Secondly,taking the 1/8 area division searches method according to the best predictive motion vector dropping range.And thirdly,basing on the distribution characteristics of the motion vectors,a searching sequence from inward to outward was taken,which can save the motion estimation time in the 5×5 search model.The experimental results shows that the proposed solution can save 33.48% motion estimation time and about27.05% in the slow motion situition in comparison with the Umhexagon S algorithm,and the PSNR and the bit rate are essentially unchanged.In order to verify the effect of the improved algorithm on the embedded system.This paper builds an embedded platform verification system based on Loongson 3A,which mainly includes software environment building based on Loongson 3A and building of Web server.The system test results show that the improved algorithm has greatly reduced the encoding time,and the quality of the test frames with low detail is basically invisible to naked eyes.