立体视频编码标准3D-HEVC的快速算法

发布时间：2018-01-21 15:59

本文关键词： 3D-HEVC 帧内预测帧间预测深度图编码快速算法　出处：《华侨大学》2017年硕士论文　论文类型：学位论文

【摘要】：为了提高3D视频的编码性能,立体视频编码联合组制定了3D-HEVC编码标准。3D-HEVC除了沿用与HEVC类似的混合编码框架和技术外,还加入了如视点间预测、深度图编码等立体视频编码新技术。与以往的基于H.264的立体视频编码标准比较,3D-HEVC具有更高的编码性能。但是,优秀的编码性能是用巨大地计算复杂度换取的,而计算复杂度的提高严重限制了3D-HEVC的实际应用。因此,研究如何降低3D-HEVC编码过程的计算复杂度,是一个重要研究课题。本文通过对3D-HEVC的编码框架以及其中的关键编码技术进行分析,对其中复杂度较高的帧内预测过程和帧间预测过程进行了一系列的优化,从而实现降低3D-HEVC编码过程的计算复杂度的目的。在帧内预测优化方面,本文提出了两种算法。首先,通过分析深度视频序列编码块的平滑程度与该编码块的最佳预测模式之间的关系,提出一种深度序列帧内快速模式选择算法SOG-FIMD(Sum-of-Gradient Based Fast Intra Depth Mode Decision)。当编码块足够平滑时,便可以跳过对一些不必要模式的计算,因为这些模式在编码块平滑时几乎不会被取到。为了判断编码块是否平滑,本文采用了计算Roberts梯度和的方式进行判定,如果梯度和小于某一阈值,则认为该编码块平滑。实验结果表明:与原始算法相比,本算法平均减少7.44%的编码时间,而BD-rate只增加0.01%。其次,通过对编码单元分割过程中采用的率失真优化函数的分析,提出一种深度序列编码单元快速分割算法SOG-FCUP(Sum-of-Gradient Based Fast CU Partition)。对于位于完全平坦区域的深度图编码块,直接终止其进一步的分割,从而降低编码器的计算开销。实验结果表明:本算法平均减少16.17%的编码时间,并且合成视点的BD-rate减少0.01%。在帧间预测优化方面,本文提出了两种算法。首先,通过对立体视频帧间预测模式分布特性进行分析,提出一种帧间预测Skip模式提前判决算法DSMD(Dependent View Skip Mode Early Determination)。利用视点相关性,根据独立视点中对应位置编码块和相邻编码块的最佳预测模式提前判决非独立视点中编码块的预测模式。实验结果表明:本算法平均减少6.50%的编码时间,而合成视点的BD-rate只增加0.08%。其次,利用视频纹理特性与预测模式之间的关系,提出一种帧间预测对称运动分割和非对称运动分割模式选择性跳过算法SASK(SMPAMP Mode Selectively Skip)。实验结果表明:本算法平均减少12.71%的编码时间,而合成视点的BD-rate仅增加0.10%。综上所述,本文所提出的优化算法能够减少3D视频编码过程的计算开销,同时保持优异的编码性能。
[Abstract]:In order to improve the performance of 3D video coding, the 3D-HEVC coding standard. 3D-HEVC is developed by the stereo video coding group. Besides, the hybrid coding framework and technology similar to HEVC are used. New stereo video coding techniques, such as interview prediction, depth map coding and so on, have been added. Compared with the previous stereo video coding standard based on H.264, 3D-HEVC has higher coding performance. Excellent coding performance is exchanged for huge computational complexity, and the increase of computational complexity seriously limits the practical application of 3D-HEVC. It is an important research topic to study how to reduce the computational complexity of 3D-HEVC coding process. This paper analyzes the coding framework and key coding technology of 3D-HEVC. In order to reduce the computational complexity of 3D-HEVC coding process, the intra prediction process and inter prediction process with high complexity are optimized. In this paper, two algorithms are proposed. Firstly, the relationship between the smoothness of the deep video sequence coding block and the best prediction mode of the block is analyzed. A fast mode selection algorithm SOG-FIMD (SOG-FIMD) is proposed for depth sequence intra frame selection (SOG-FIMD). Sum-of-Gradient Based Fast Intra Depth Mode decision. When the encoding block is smooth enough. It is possible to skip the calculation of unnecessary patterns, which are rarely taken when the encoding block is smooth, in order to determine whether the block is smooth or not. In this paper, the Roberts gradient sum is calculated. If the gradient sum is less than a threshold, the coding block is considered smooth. The experimental results show that the proposed algorithm is compared with the original algorithm. The average coding time of this algorithm is reduced by 7.44%, while the BD-rate is only increased by 0.01. Secondly, the rate-distortion optimization function used in the segmentation of the coding unit is analyzed. A Fast Segmentation algorithm SOG-FCUP(Sum-of-Gradient Based Fast CU Partition for Deep sequence coding Unit. For a depth map block located in a completely flat region. The further segmentation is directly terminated to reduce the computational cost of the encoder. The experimental results show that the algorithm reduces the coding time by 16.17% on average. In the aspect of inter-frame prediction optimization, two algorithms are proposed in this paper. Firstly, the characteristics of prediction mode distribution between stereo video frames are analyzed. DSMD(Dependent View Skip Mode Early determination algorithm for interframe predictive Skip mode is proposed. Make use of the relevance of the viewpoint. According to the best prediction mode of corresponding position coding block and adjacent coding block in independent view, the prediction mode of coding block in independent view is determined in advance. The experimental results show that:. The average coding time of this algorithm is reduced by 6.50%. However, the BD-rate of synthetic view only increases by 0.08. Secondly, the relationship between video texture characteristics and prediction mode is used. A selective skipping algorithm SASK(SMPAMP Mode Selectively Skip-Predictive symmetric Motion Segmentation and asymmetric Motion Segmentation Mode between frames is proposed. The experimental results show that the algorithm reduces the coding time by 12.71% on average. In conclusion, the proposed optimization algorithm can reduce the computational overhead of 3D video coding process while maintaining excellent coding performance.
【学位授予单位】：华侨大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN919.81

【相似文献】