基于GPU的H.264并行解码优化

Parallel decoding optimization of H.264 based on GPU

摘要:

H.264视频编码标准因其很好的压缩率而成为目前的主流标准之一；针对H.264解码复杂度提高、计算量增大的现状，根据GPU适合通用并行计算的特性，提出其基于GPU的并行解码优化。使用GPU对帧内预测与滤波器模块解码，CPU负责控制GPU以及对剩余部分解码。通过对帧内预测解码的分析，提出一种优化的帧内预测并行算法，经实验证明相比有优化前算法解码效率被提高20%；通过对滤波器模块的研究，提出一种滤波强度并行求取算法以及并行滤波执行算法，经实验证明滤波器的处理速度提升了30%，且相比原图像△PSNR最大为0.10，△SSIM为0.01。最终通过实验证明，使用GPU对视频解码的关键模块处理，能大大提高处理效率。

Abstract:

H.264 video coding standard has become one of the mainstream standards because of its good compression ratio. In order to deal with the increasing complexity and computation in H.264 decoding, it is proposed in this paper that the parallel decoding optimization based on GPU which has the characteristics suitable for general parallel computing. Using GPU to decode intra prediction and filter modules, CPU controls GPU and decodes the remaining part. Based on the analysis of intra prediction decoding, a parallel intra prediction algorithm is also proposed, the experimental results of which show that the intra prediction decoding efficiency is increased by 20%. Through the study of the filter module, a parallel filtering algorithm and parallel filtering execution algorithm are proposed. It is proved by experiments the processing speed of the filter is increased by 30%, and the maximum ΔPSNR is 0.10 compared with the original image, and the ΔSSIM is 0.01. Finally, it is proved by experiments that using GPU to process the key modules of video decoding can greatly improve the processing efficiency.