基于时空特征的HEVC压缩视频差错掩盖算法研究

发布时间：2018-03-13 11:07

本文选题：HEVC　切入点：差错掩盖(EC)　出处：《南京邮电大学》2017年硕士论文　论文类型：学位论文

【摘要】：高效视频编码(HEVC)是最新一代视频编码标准,能够以一半的码率获得与H.264/AVC标准相同的编码质量。但是压缩率的提升,使得视频码流对传输差错更加敏感,必须采取有效的差错控制措施。基于解码端的差错掩盖算法具有突出的优点,一直是差错控制算法的研究热点。本文针对差错掩盖算法进行了研究,结合三种不同的差错状况,发掘HEVC标准的编码特点,分别提出有效的差错掩盖算法。主要工作及创新点如下:(1)对于HEVC码流整帧丢失的情况,针对已有差错掩盖算法的不足提出了基于PU的运动矢量外推差错掩盖算法。首先建立HEVC标准下基于预测单元(PU)的MVE算法;然后对丢失帧进行PU矢量外推,根据最小尺寸的外推PU块,推测LCU的分割模式,由此确定当前待掩盖LCU的差错掩盖基本单元(ECBU);接着对每个ECBU进行基于PU的MVE掩盖,从而完成该LCU的掩盖。依次掩盖每一个LCU,直至掩盖完整个丢失帧。(2)对于HEVC码流LCU丢失的情况,提出基于PU融合的加权BMA掩盖算法。该算法首先根据相邻帧之间的高度时域相关性,将丢失LCU的CU/PU分割模式假定为和其同位块的相同;接着针对由率失真代价选择出的运动矢量不一定代表视频真正的运动的问题,根据同位块残差能量判断当前的分割模式是否可靠,并将相邻的不可靠PU进行合理融合,完成对丢失LCU分割模式的优化;然后是针对在HEVC标准下直接使用BMA算法的不足,建立权值图进行基于PU的BMA掩盖,每掩盖完一个PU都要更新一次权值图,如此反复进行直至掩盖完整个丢失LCU。(3)鉴于在差错掩盖算法中考虑视频内容特性的必要性,提出基于视频内容特性判断的HEVC掩盖算法。首先通过分析HEVC标准编码树单元分割为编码单元(CU)时的特性,得出CU划分深度与图像的不同内容分布有着密切的关系,可以由CU深度信息对CU所在图像区域的内容特征做出推断——丢失LCU的深度由其空时域周围正确接收LCU的编码深度信息来推测。如果判定丢失LCU所在图像区域纹理简单、运动平缓,该丢失LCU应用简便的掩盖方法——运动矢量由相邻正确接收编码单元的运动矢量求均值得到,提高掩盖效率。另外根据HEVC标准将CU编码为帧内模式时的特性,通过计算丢失LCU周围正确接收LCU编码为帧内模式的CU数量占总CU数量的比率,推测出该丢失LCU所在图像区域是否属于运动非常剧烈区域,如果属于,应该使用空域掩盖算法,本文结合HEVC帧内纹理角度预测技术提出了基于帧内纹理角度模式的空域掩盖算法;剩下图像区域的内容特性则属于一般情况,采用基于PU融合的差错掩盖算法就可以取得良好的掩盖效果。论文最后对全文工作进行了总结,并对后续研究方向进行了展望。
[Abstract]:High efficiency video coding (HEVC) is the latest generation video coding standard, which can achieve the same coding quality as H.264 / AVC standard at half bit rate. However, the improvement of compression ratio makes video stream more sensitive to transmission errors. It is necessary to take effective error control measures. The error concealment algorithm based on decoding has outstanding advantages and has always been the research hotspot of error control algorithm. In this paper, the error concealment algorithm is studied and combined with three different error situations. This paper explores the coding characteristics of HEVC standard and proposes effective error concealment algorithms respectively. The main work and innovation are as follows: 1) for the whole frame loss of HEVC bitstream, A motion vector extrapolation error concealment algorithm based on pu is proposed to overcome the shortcomings of existing error concealment algorithms. Firstly, the MVE algorithm based on prediction unit is established in HEVC standard, and then the lost frame is extrapolated by pu vector. According to the minimum size extrapolation pu block, the partition mode of LCU is inferred, and then the error concealment basic unit of current LCU is determined, and then each ECBU is masked by MVE based on pu. This masked the LCU. Each LCU is masked in turn until the entire lost frame. 2) for the HEVC bitstream LCU loss, A weighted BMA masking algorithm based on pu fusion is proposed. Firstly, according to the high temporal correlation between adjacent frames, the CU/PU segmentation mode of lost LCU is assumed to be the same as that of the same block. Then, aiming at the problem that the motion vector selected by rate-distortion cost does not necessarily represent the real motion of video, judging whether the current segmentation mode is reliable or not according to the residual energy of the same block, and merging the adjacent unreliable pu reasonably. The optimization of the missing LCU segmentation mode is completed, and then the weight map is established to cover the BMA based on pu, aiming at the deficiency of using the BMA algorithm directly under the HEVC standard. Every time a pu is covered up, the weight value graph is updated once. In view of the need to consider video content characteristics in error concealment algorithms, This paper proposes a HEVC concealment algorithm based on video content characteristic judgment. Firstly, by analyzing the characteristics of the HEVC standard coding tree unit when it is divided into the encoding unit (CUU), it is concluded that the depth of the CU partition is closely related to the different content distribution of the image. CU depth information can be used to infer the content features of the image region in which the CU is located-the depth of the missing LCU can be inferred by correctly receiving the encoding depth information of the LCU around its space-time domain. If it is simple to determine the texture of the image region in which the LCU is lost, The lost LCU uses a simple concealment method, the motion vector is obtained from the mean value of the motion vector of the adjacent correct receiving coding unit, which improves the concealment efficiency. In addition, according to the HEVC standard, the CU is encoded into the intra-frame mode. By calculating the ratio of the number of CU encoded by the received LCU to the total number of CUs around the lost LCU, it is inferred whether the image region of the lost LCU belongs to the highly moving region. If so, the spatial domain masking algorithm should be used. In this paper, a spatial domain masking algorithm based on intra-frame texture angle pattern is proposed based on HEVC intra texture angle prediction technology, and the content characteristics of the remaining image regions are general. The error concealment algorithm based on pu fusion can achieve good concealment effect. Finally, the paper summarizes the work of this paper, and looks forward to the future research direction.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN919.81

【参考文献】