基于三步快速选择的HEVC帧内预测优化算法

发布时间：2018-03-26 14:23

本文选题：HEVC　切入点：帧内预测　出处：《西安电子科技大学》2014年硕士论文

【摘要】：HEVC(High Efficiency Video Coding)是继H.264/AVC之后的新一代视频编码标准,由协作视频编码组JCT-VC联合制定。HEVC保留了基于块运动补偿的传统混合视频编码框架,同时在帧内预测、帧间预测、正交变换与量化、熵编码等关键模块部分都采用了相应的新技术,使得HEVC在基本保证H.264/AVC所有的功能与应用的情况下,其编码效率有了显著提升,相较于H.264/AVC提高了一倍。HEVC优异的编码性能是以较大的编码复杂度为代价实现的,比如四叉树递归块划分结构、率失真优化决策、35种帧内预测方向等关键技术的复杂度都非常高。本文对HEVC帧内编码进行了深入研究,降低帧内编码复杂度,提高整体编码效率。针对帧内预测过程中的模式选择及影响模式选择性能的相关部分,从帧内分块的四叉树遍历结构和帧内预测模式的选择过程两方面进行了深入分析,并对帧内预测各部分的计算量进行了统计。对于LCU其帧内编码存在的可能划分方式有1024种,帧内预测多达35个,对35种模式都要进行率失真优化决策,由此导致帧内预测运算量非常庞大。本文针对HEVC帧内预测瓶颈问题,提出了一种基于三步快速选择的HEVC帧内预测优化算法。帧内预测最佳模式选择需两个基本步骤:粗略模式选择(RMD过程)和模式细选(RDO过程)。针对两个过程分别进行详细分析,提出了相应的优化算法主要策略如下:(1)基于三步选择的粗选优化算法。本算法在模式粗略选择(RMD,Rough Mode Decision)过程中优先搜索预设大方向、再在选定大方向范围内精确搜索两步,使得原需遍历多达35种帧内预测模式,减少到最多遍历15种模式即能得出最佳预测模式,以此减少了一半的需计算SATD代价的模式个数;(2)基于视频特性的细选优化算法。本算法利用最有可能模式(MPM,Most Probable Mode)和第一个候选模式的相关性,提出了最佳预测模式选择的最后一步,根据PU尺寸的大小自适应地减少了参与率失真代价计算的模式个数,有效减少细选计算量。为验证上述三步快速选择算法的效率,本文选择HEVC标准测试序列对其进行了系统的测试。本文提出的优化算法与标准算法相比较,实验结果显示:在不同序列、不同QP参数的配置情况下,本算法保证视频质量基本不变,码率稍有增加,平均在0.2%左右;亮度峰值信噪比PSNR的降低量均在0.05d B左右,保证在0.1dB之内;而编码时间平均减少了约40%,很大程度上加快了编码速度。实验结果可以充分表明,本文实现的算法能够有效的提高编码效率、缩短帧内编码时间,对HEVC的实际应用具有积极的意义。
[Abstract]:HEVC(High Efficiency Video coding is a new generation video coding standard after H.264/AVC. It is developed by cooperative video coding group (JCT-VC). It preserves the traditional hybrid video coding framework based on block motion compensation, and at the same time, it is used in intra-frame prediction, inter-frame prediction, quadrature transform and quantization. Some key modules, such as entropy coding, have adopted the corresponding new technology, which makes the coding efficiency of HEVC improve significantly when all functions and applications of H.264/AVC are basically guaranteed. Compared with H.264/AVC, the excellent coding performance of HEVC is achieved at the cost of large coding complexity, such as quadtree recursive block partition structure. The complexity of the key techniques such as rate-distortion optimization decision and 35 kinds of intra prediction directions are very high. In this paper, the intra coding of HEVC is studied in depth to reduce the intra coding complexity. In order to improve the overall coding efficiency, the mode selection in the intra prediction process and the related parts that affect the performance of the mode selection are analyzed in detail from two aspects: the quadtree traversal structure of the intra block and the selection process of the intra prediction mode. For LCU, there are 1024 possible partition modes, 35 intra prediction modes, and rate-distortion optimization decision for all 35 modes. As a result, the intra prediction computation is very large. This paper aims at the bottleneck problem of intra prediction in HEVC. This paper presents an algorithm for intra-frame prediction optimization of HEVC based on three-step fast selection. It requires two basic steps: rough mode selection (rough mode selection) and mode selection process (RDO). The two processes are analyzed in detail. The main strategy of the corresponding optimization algorithm is as follows: 1) rough selection algorithm based on three-step selection. In the process of rough pattern selection, the algorithm gives priority to searching for the default direction, and then accurately searches two steps within the selected broad direction. This makes it possible to traverse as many as 35 intra prediction modes, so that the best prediction mode can be obtained by reducing the traversal to up to 15 modes. In this way, the number of modes needed to calculate the cost of SATD is reduced by half.) the algorithm is based on video features. This algorithm utilizes the correlation between the most probable mode and the first candidate mode. The last step of optimal prediction mode selection is proposed. According to the size of pu, the number of modes calculated by participating rate-distortion cost can be reduced adaptively, and the computation of fine selection can be effectively reduced. In order to verify the efficiency of the three-step fast selection algorithm mentioned above, In this paper, the HEVC standard test sequence is selected to test it systematically. Compared with the standard algorithm, the experimental results show that: under different sequences and different QP parameters configuration, The algorithm ensures that the video quality is basically unchanged, the bit rate is slightly increased, the average is about 0.2%, and the decrease of PSNR is about 0.05 dB, which is within the 0.1dB. However, the average coding time is reduced by about 40 percent, which greatly accelerates the coding speed. The experimental results show that the proposed algorithm can effectively improve the coding efficiency and shorten the coding time within frames. It is of positive significance to the practical application of HEVC.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN919.81

【参考文献】