Research On Key Technology Of H.265/HEVC Encoder Based On CPU+GPU Heterogeneous Platform

In recent years,with the rapid development of economy,society and technology,the 4K HD video is coming into people's life step by step.Compared with the currently popularized 1080p HD video,the 4K HD video reaches 8 times the amount of data,but the transmission bandwidth has not increased significantly.Although the H.265/HEVC coding standard for ultra high definition video compression has been published for 5 years,it has not been popularized and popularized in application because of its high computational complexity.At present,the optimization of H.265/HEVC encoder is mainly focused on the optimization of mode fast selection algorithm and the design of multi thread parallel mode.The optimization performance can not meet the practical application requirements of 4K ultra high definition video compression in the field of radio and television.Therefore,based on the heterogeneous platform of CPU+GPU,this paper studies the time-consuming intra prediction and inter prediction module of H.265/HEVC encoder and the load balancing of band level parallel coding,and improves the coding speed of H.265/HEVC on the premise of guaranteeing the coding quality.The main work of this article is as follows:(1)Aiming at the problem that the reconstructed pixels can not be obtained in GPU intra prediction,a hierarchical intra prediction method is proposed,that is,the prediction is started from the smallest code block to the maximum code block,and the prediction of each layer can make use of the rebuilt pixels generated by the next layer.In addition,the prediction process of each coding block is optimized,including the combination of reference pixel generation and reference pixel filter module,the pre selection of intra prediction mode based on the gradient matrix,the pixel prediction in the frame prediction mode and the encoding bit estimation method at the GPU end.(2)The problem that the global optimal solution can not be obtained by the parallel fast search algorithm can not be obtained.By using the fitness of the full search of the motion estimation,the large block of SA is synthesized by the small block of SAD.D can reduce the computation cost of large SAD while improving coding performance.(3)Using the predicted data returned by GPU,the fast depth decision and fast mode selection algorithm for the CPU end are designed,and a coding time prediction method based on the coding parameters is established to solve the problem of load imbalance in H.265/HEVC band level parallel.In the process,we dynamically modify the weight of each reference frame to achieve HE VC dynamic load balancing.Finally,based on the above work,we implement a H.265/HEVC encoder based on CPU+GPU heterogeneous platform.By comprehensively utilizing the advantages of CPU and GPU to optimize the H.265/HEVC encoder,the encoding complexity of the encoder is greatly reduced,and the coding efficiency is improved.The experimental results show that the optimization of H.265/HEVC encoder based on CPU+GPU isomeric platform is a feasible and effective method to reduce the computational complexity of H.265/HEVC encoder,which is of great significance to the development of 4K TV.