新型视频编码快速算法研究

发布时间：2018-07-10 13:18

本文选题：高效视频编码 + 帧间预测　；参考：《华侨大学》2017年硕士论文

【摘要】：随着高清视频与超高清视频的普及,为了方便视频在有限的网络带宽上传输,视频编码联合小组在2013年确定了高效视频编码作为新一代的视频编码标准。它是在上一代视频编码标准H.264的基础上进行了大量的改进与创新,HEVC采用了更大尺寸的块结构与更多的预测模式,有效地提升了压缩率,同时带来的计算复杂度也成倍的增加。本文首先介绍HEVC的编解码框架与引入的主要技术点,重点介绍了帧内预测过程与帧间预测过程。随后针对帧间预测过程引入的巨额的计算复杂度,提出了多种算法降低帧间预测过程的计算复杂度。本文对帧间层面的优化分为两个方面进行:宏观级和微观级。在宏观级层面的优化上:1.根据图像块之间的空域相关性,利用空间邻近编码树单元的最大深度来预测当前编码树单元的深度范围;2.统计分析编码单元选择分割的率失真代价与不分割的率失真代价取值状况,提出一种软阈值法来终止编码单元继续分割。实验结果表明,本文提出的帧间宏观级算法可以减少33%的编码时间,平均码率差增加了0.454%,平均峰值信噪比差降低了0.017dB。在微观级层面的优化上:1.利用运动矢量估计图像块的运动状况,决策跳过部分预测模式;2.分析残差块的纹理分布,针对不同的纹理分布对预测模式选择过程做进一步的细化处理;3.利用在检测每一预测模式时,计算变换单元的纹理复杂程度,提前决策变换单元是否继续分割。实验结果表明,本文提出的帧间微观级算法减少50.6%的编码时间,平均码率差增加了1.606%,平均峰值信噪比差降低了0.06dB。将宏观级算法和微观级算法综合起来得到帧间综合算法,实验结果表明,帧间综合算法可以减少57.1%的编码时间,平均码率差增加了1.994%,平均峰值信噪比差降低了0.074dB。鉴于综合算法引入了一定程度的率失真性能的损失,随后提出了以自由地控制错检概率,尽可能地降低漏检概率的决策算法--Neyman-Pearson模式决策算法。该算法在线统计训练数据,并周期性地更新训练数据,通过训练数据来提前决策skip模式为最优预测模式与提前终止编码单元继续分割。实验结果表明,Neyman-Pearson模式决策算法能够减少39.0%的计算复杂度,平均码率差增加了0.782%,平均峰值信噪比差降低了0.027dB。在论文的最后,总结了本文所做的研究工作,同时提出了未来的研究方向。
[Abstract]:With the popularity of high-definition video and ultra high definition video, in order to facilitate video transmission on limited network bandwidth, the video coding joint team identified efficient video coding as a new generation of video coding standards in 2013. It has made a lot of improvements and innovations on the basis of the previous generation of video coding standard H.264, and HEVC adopted more. The large size block structure and more prediction mode can effectively increase the compression rate and increase the computational complexity. This paper first introduces the coding and decoding framework of HEVC and the main technical points introduced, focusing on the intra prediction process and inter frame prediction process. In this paper, the optimization of inter frame level is divided into two aspects: macro level and micro level. At the macro level, 1. according to the spatial correlation between the image blocks, the current coding tree is predicted by the maximum depth of the space adjacent coding tree unit. The depth range of the unit; 2. a statistical analysis of the rate distortion cost and the undivided rate distortion cost of the selected segment of the coding unit. A soft threshold method is proposed to terminate the continuous segmentation of the coding unit. The experimental results show that the inter frame macro level algorithm proposed in this paper can reduce the encoding time by 33%, the average bit rate difference increases by 0.454%, and the average code rate difference is increased by 0.454%. The difference of mean peak signal to noise ratio (SNR) reduces the optimization of 0.017dB. at the micro level: 1. using motion vectors to estimate the motion situation of the image block, the decision skip part of the prediction mode; 2. analyze the texture distribution of the residual block, make a step of refining the prediction mode selection process for different texture distribution, and 3. use to detect each prediction model. At the same time, the texture complexity of the transformation unit is calculated and whether the transformation unit will continue to be divided in advance. The experimental results show that the proposed inter frame micro level algorithm reduces the encoding time by 50.6%, the average bit rate difference is increased by 1.606%, the average peak signal to noise ratio difference is reduced by 0.06dB., and the macro level algorithm and the micro level algorithm are synthesized. The experimental results show that the inter frame synthesis algorithm can reduce the encoding time by 57.1%, the average bit rate difference is increased by 1.994%, the average peak signal to noise ratio difference is reduced by 0.074dB. in view of the loss of a certain degree of rate distortion introduced by the integrated algorithm, and then it is proposed to control the misdetection probability freely and reduce the leakage as much as possible. The --Neyman-Pearson model decision algorithm of probabilistic decision algorithm. This algorithm trains data online, periodically updates the training data, and makes the skip mode ahead of time by training data to continue the segmentation of the optimal prediction model and the advance termination coding unit. The experimental results show that the Neyman-Pearson pattern decision algorithm can reduce 39%. The calculation complexity, the average bit rate difference increased by 0.782%, the average peak signal to noise ratio difference reduced 0.027dB. at the end of the paper, summed up the research work done in this paper, and put forward the future research direction.
【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】