Research And Implementation Of H.265 Encoding Technology Based On GPU

With the increasing scope of video applications,video coding standards are constantly being updated.The constant change of video formats leads to an increase in the amount of video data.Therefore,a good video compression standard is particularly important.The newgeneration video coding standard H.265/HEVC can reduce the code rate by 50% compared with the previous H.264/AVC coding standard while guaranteeing the video coding quality,but its complexity is also greatly increased.Slow coding.The H.265 official open-source software-encoder HM uses a serial implementation.The encoding speed is very slow but there is no copyright protection and it is convenient to develop.The H.265 encoding technology also includes a large number of repetitive calculations and parallel modules.At the same time,the GPU is a highly parallel architecture graphics card,which is very suitable for parallel computing.The combination of the two will increase the encoding speed.In this paper,the H.265 encoding technology is designed and implemented in parallel on the GPU platform.The CUDA language is used to rewrite the time-consuming module in the HM encoder to achieve the purpose of improving the encoding speed.First of all,this paper analyzes the key technologies of H.265 coding standard and coding in detail,including the image partitioning method,the overall coding framework,intra prediction,inter-frame prediction,transformation and quantization,and details the GPU architecture and CUDA architecture.The narrative lays a theoretical foundation for subsequent implementation on the GPU platform.The time-consuming proportion of the HM encoder's operation framework and each encoding module is analyzed,and the interframe prediction,transformation and quantization are selected as the encoding module implemented in parallel in this paper.Then,in view of the high complexity of intra-prediction mode selection algorithms,this paper proposes a method for fast selection of intra-prediction modes.It divides the region by 33 types of angular prediction modes and uses the high-probability prediction mode in the divided regions.Continuous matching,quick positioning of the optimal prediction mode distribution,reducing the number of optimal prediction modes and reducing the complexity of the mode selection,without affecting the encoding quality.In the inter-frame prediction parallel,this article chooses the prediction accuracy.The highest full search algorithm overcomes the disadvantages of the large amount of computation in the algorithm by virtue of the parallelism of the GPU.In the SAD calculation section,the calculation of all the size blocks is completed by a 4×4 block step-by-step superposition method.The mapping function is designed to display the memory.The memory address corresponds to the block size.In the transformation and quantization module,parallel parallelism and column parallelism are used in this paper to complete the parallel design and implementation,which effectively improves the encoding speed.Finally,this paper tests the parallel designed H.265 encoder on the GPU(Ge Force GTX 1080)platform,compares it with the HM encoder under the CPU platform,and optimizes the selection of the intra prediction mode.Test verification,independent time-consuming analysis and comparison of inter prediction,transformation and quantization,respectively,verifies that the encoder can correctly encode video sequences,and the overall encoding speed is increased by an average of 9 times under the premise of ensuring the encoding quality.The speedup.In the field of H.265 coding,especially in the case of high coding performance requirements have application prospects.