HEVC帧内CU划分快速算法研究

Research on Fast CU Partition Algorithm for HEVC Intra Prediction

为了降低帧内编码单元(CU)四叉树划分的计算复杂度,本文提出了一种基于多重纹理的帧内CU划分快速算法。通过提取CU的内容复杂度、方向复杂度以及子CU的内容复杂度差异和方向复杂度差异作为纹理特征,设计了基于双SVM的CU划分决策模型,实现了对复杂纹理CU的提前划分和简单纹理CU的提前终止划分。实验结果表明,该快速算法与x265-1.7相比,码率增加0.47%,亮度峰值信噪比降低0.012 d B,编码时间减少36.87%,有效提高了帧内预测效率。

In order to reduce the computational complexity of Intra Coding Unit （CU） quadtrcc partitioning, a novel intra CU partitioning algorithm based on multiple textures is proposed in this paper. By extracting the content complexity and direction complexity of the CU and the differences in content complexity and direction complexity of the sub - CU as texture teatures, a CU paltition decision model based on double SVM is designed, which implements the early split of CUs with complex texture and early termination of CUs with simple texture. Experimental results show that compared with x265 - 1.7, the proposed fast algorithm can save 36.87% encoding time with 0.47% of bitrate increment and 0. 012 dB loss of peak signal to noise ration of luminance, which effectively improves efficiency of the inlra prediction,