Research On The Optimization Of AVS And HEVC Encoders

With the rapid development of Internet video industry,the pressure of video content storage and transmission is increasing.Thus efficient video compression technology is particularly important.At present,the latest generation of the international video coding standard is HEVC,while the domestic AVS series coding standard with independent technical property has been developed to AVS2.However,due to the use of more sophisticated coding technology,new video coding standards have high coding complexity,which limits their popularity in practical applications.In this regard,it is necessary to optimize the video encoder to improve the coding speed while maintaining high compression performance.The optimization of the encoder usually consists of two directions: one is optimization of engineering implementation and the other is algorithm optimization,which complement each other.In this paper,we explore the optimization methods of the encoder on multiple video coding standards.Firstly,we propose a comprehensive optimization method based on Skip decision process and subjective video coding for realtime low bitrate condition,which greatly improves the coding performance while keeping the real-time encoding speed of xAVS encoder.Secondly,we accelerated the computation intensive modules in the AVS2 encoder using the latest SIMD instruction set AVX2,including intra prediction,transform,subpixel interpolation and dual satd.Without losing any coding performance,we greatly improved the computation speed of these modules.Finally,aiming at the CU partition decision in HEVC intra coding,a fast algorithm based on convolutional neural network is proposed.A convolutional neural network is designed to predict the whole partition of a LCU,which then helps to guide the fast encoding process.Compared with the original HM encoder,the algorithm can save 58.35% coding time while the average BDBR increase is 1.509%.