摘要： 针对高效视频编码(HEVC)中编码单元(CU)进行四叉树递归遍历时间较长的问题,提出一种基于纹理四叉树的快速HEVC帧内编码算法。采用Sobel算子对当前视频图像提取边缘,利用最大类间方差法剔除弱边缘并保留强边缘。通过递归方式对边缘图中的每个64×64单元建立纹理四叉树,使用视频图像的纹理四叉树结构对CU最优分割组合进行预测。对于不同大小的分割单元,无需完全递归遍历所有的CU深度,从而缩小CU搜索范围,节省编码时间。实验结果表明,与HEVC标准算法相比,该算法亮度分量的码率平均增加了0.50%,信噪比和编码时间分别减少了0.03 dB和28.70%。

关键词: 高效视频编码, 帧内预测, 快速编码, 纹理四叉树, 边缘提取, 编码单元深度

Abstract: Aiming at high computing complexity of recursive quadtree partitioning of a large Coding Unit(CU) in High Efficiency Video Coding(HEVC),a fast intra coding algorithm of HEVC based on texture quadtree is proposed in this paper.The edges of the whole frame are extracted by Sobel operator.The weak edges are removed and the edge map is formed using maximum between-cluster variance method.Then,the texture quadtree is recursively built for each 64×64 of edge map.The texture quad-tree is applied to predict the CU partition combination.For different partition size,full searching all CU depth is avoided.Hence,CU search range is narrowed and the encoding time is reduced.Experimental result shows that the proposed algorithm saves 28.70% of the encoding time on average,while it only leads to increase 0.50% Bjontegaard Delta Bit Rate(BDBR) and decrease 0.03 dB Bjontegaard Delta Peak Signal-to-Noise Ratio(BDPSNR) for luminance component.

Key words: High Efficiency Video Coding(HEVC), intra prediction, fast coding, texture quadtree, edge extraction, Coding Unit(CU) depth

中图分类号: 

• TP391