面向3D-HEVC深度图编码的快速优化算法研究

发布时间：2018-11-09 16:29

【摘要】：随着多媒体通信等技术和各种视频终端处理能力的快速发展,3D视频越来越在生活中普及应用。上一代基于H.264的多视点视频编码标准不能满足当前与日俱增的3D视频数据量的高效压缩,因此,立体视频编码联合组(The Joint Collaborative Team on 3D Video Coding Extension Development,JCT-3V)制定了新一代多视点视频编码标准 3D-HEVC(3D-High Efficiency Video Coding)。尽管3D-HEVC取得较高的编码效率,但是其存在计算复杂度较高的问题,严重限制了 3D视频的实际应用。因此,如何在保证编码3D视频质量的前提下,最大幅度的降低3D-HEVC的计算复杂度,成为当前视频技术领域的一个研究热点。目前通用的3D视频格式为纹理视频加深度视频格式(Multi-view Video Plus Depth,MVD),其中深度图对虚拟视点合成具有重要作用,然而也引入了较大的计算复杂度。为此,本文针对3D-HEVC深度图编码,深入分析了深度图特性,提出一系列快速优化算法,以保持编码性能的前提下极大的节省计算复杂度。本文的主要工作如下:1、提出一种基于深度分类的低复杂度深度图帧内预测算法。本算法根据深度图特征把3D-HEVC帧内预测模式分成三类——平滑类,方向角类和深度类。首先采用HOG特征对深度预测块进行特征提取,其次用SVM训练器对所提取的特征进行模式判决,根据判决结果对深度块所属类型中的所有模式进行RD-Cost计算得到最佳的预测模式。实验结果表明,与原有3D-HEVC相比,本算法平均缩短34.85%的编码时间,而BD-Rate仅降低了 0.14%。2、提出了一种加速深度块CU的分割方法。原有3D-HEVC采用递归的方法分割CU,消耗大量的编码时间。对此,本文提出了一种快速终结CU递归划分的方法。首先计算当前编码CU的方差和对角像素差的绝对值之和,通过阈值法比较,判定是否需要提前终止CU的划分。本算法的实验结果表明,与原始3D-HEVC相比,本算法平均减少了 9.73%的编码时间,而BD-Rate仅升高了 0.02%。3、最后,本文针对SDC编码进行优化。通过实验统计发现,SDC编码的选的与当前预测单元PU的平滑性息息相关,若PU比较平滑,则选择SDC编码的可能性较高。因此,本算法在得到全搜索列表后,先计算当前PU外圈像素差的绝对值之和,进行阈值化比较,判定是否跳过非SDC编码,以降低计算复杂度。实验结果表明,该算法平均能减少10.64%的编码时间,而仅造成0.16%BD-Rate的增加,此外本文最后将所提3种算法进行整合,系统的优化深度图的帧内预测过程。实验结果表明,与原有3D-HEVC相比,本算法能减少的编码时间平均能达到43.09%,而BD-Rate仅增加了 1.06%。综上,本文围绕3D-HEVC深度图编码提出了一系列优化算法,在保持3D-HEVC编码效率的前提下,有效的降低了计算复杂度。本文的研究成果对于促进3D-HEVC的应用具有一定的意义和价值。
[Abstract]:With the rapid development of multimedia communication technology and various video terminal processing capabilities, 3D video has become more and more popular in life. The previous generation of the H.264-based multi-view video coding standard does not meet the high-efficiency compression of the current increasing amount of 3D video data. Therefore, the Joint Collaborative Team on 3D Video Coding Extension Development (JCT-3V) has developed a new-generation multi-view video coding standard, 3D-High Efficiency Video Coding. Although the 3D-HEVC has higher coding efficiency, it has a high computational complexity and severely limits the practical application of 3D video. Therefore, how to reduce the computational complexity of the 3D-HEVC on the premise of ensuring the quality of the 3D video is a hot topic in the current video technology field. At present, the general 3D video format is the multi-view video Plus Depth (MVD) of the texture video, in which the depth map plays an important role in the virtual viewpoint synthesis, but also introduces a large computational complexity. In this paper, the 3D-HEVC depth map is coded, the depth map is deeply analyzed, and a series of fast optimization algorithms are proposed to save the computational complexity on the premise of keeping the coding performance. The main work of this paper is as follows: 1. A low-complexity depth-map intra-frame prediction algorithm based on depth classification is proposed. In this algorithm, the 3D-HEVC intra-prediction mode is divided into three classes _ smoothing class, direction angle class and depth class according to the depth map feature. The method comprises the following steps of: firstly, carrying out feature extraction on a depth prediction block by adopting a HOG characteristic, and secondly, performing mode judgment on the extracted feature by using an SVM training device, and performing RD-Cost calculation on all modes in the type of the depth block according to the judgment result to obtain an optimal prediction mode. The experimental results show that, compared with the original 3D-HEVC, the average time of the algorithm is shortened by 34. 85%, and the BD-Rate is only reduced by 0.14%. The original 3D-HEVC divides the CU by a recursive method and consumes a large amount of encoding time. In this paper, a method for quickly terminating a CU recursive partition is presented in this paper. First, the sum of the absolute value of the variance of the current code CU and the diagonal pixel difference is calculated and compared by the threshold method, it is determined whether the division of the CU is required to be terminated in advance. The experimental results of this algorithm show that, compared with the original 3D-HEVC, the mean time of the algorithm is reduced by 9.73%, and the BD-Rate only increases by 0. 02%. 3, and finally, this paper is optimized for the SDC coding. It is found that the choice of the SDC coding is closely related to the smoothness of the current prediction unit PU, and if the PU is relatively smooth, the possibility of selecting the SDC coding is high. Therefore, after a full search list is obtained, the sum of the absolute values of the current PU outer ring pixel difference is calculated, and the threshold value comparison is performed to determine whether the non-SDC encoding is skipped to reduce the computational complexity. The experimental results show that the algorithm can reduce the coding time of 10.64%, and only result in an increase of 0. 16% BD-Rate. In addition, the three algorithms are combined to optimize the intra-frame prediction process of the depth map. The experimental results show that compared with the original 3D-HEVC, the coding time can be reduced by 43. 09%, and the BD-Rate only increases by 1.06%. In this paper, a series of optimization algorithms are proposed around the 3D-HEVC depth map coding, and the computational complexity is effectively reduced on the premise of maintaining the coding efficiency of the 3D-HEVC. The research results of this paper are of great significance and value to the application of 3D-HEVC.
【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】