HEVC帧间快速编码算法的研究

发布时间：2017-12-31 16:22

本文关键词：HEVC帧间快速编码算法的研究　出处：《电子科技大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着高清视频应用的快速发展,视频传输和储存方面产生了巨大的挑战。在这一新环境下,Joint Collaborative Team on Video Coding(JCT-VC)工作小组制定了新的视频编码标准High Efficiency Video Coding(HEVC/H.265)来应对挑战。HEVC为了达到高效编码的目的,在原有视频编码标准Advanced Video Coding(AVC/H.264)的基础上增加了许多更加先进的视频编码技术,如灵活块划分方式、非对称运动模式划分、多预测块模式。与H.264相比,HEVC在获得相同视频质量的前提下可以降低50%左右的传输比特流,但在编码复杂度计算量上也大大地增加。在计算能力一般的无线移动终端设备上,由于设备不能支持所有编码技术的使用,这势必会大大降低编码效率,从而无法突显HEVC编码效率上的优势。在视频编码技术中预测编码是计算比重最大的一部分,而其中的帧间预测编码又是重要的环节之一。因此,HEVC帧间快速编码算法研究是当今视频编码领域中非常重要的研究内容之一。本文首先简单地介绍了视频编码标准的发展历程和视频编码的整体框架,然后对当今国内外在HEVC的快速编码的研究做了系统的总结。接着,对HEVC的编解码框架和相关关键技术作了简要介绍。最后,针对HEVC帧间CU(Coding Unit)块划分和PU(Prediction Unit)预测模式做了详细说明和深入研究。在针对HEVC的CU块的划分过程中,本文先参考空域上的邻近LCU(Largest Coding Unit)的最大划分深度(Depth)为当前LCU划定一个范围,从而跳过部分深度下的PU预测模式判决过程,有效的降低编码复杂度。为了进一步降低CU划分的计算复杂度,在每个深度进行完PU模式预测判决之后,根据最佳模式下的残差信息以及变换量化系数矩阵提出了提前结束CU划分的算法。在针对HEVC帧间PU预测模式判决过程中,本文提出了针对编码过程中占比最大的两种PU模式——Skip模式和2Nx2N模式的快速算法。为了提前预测Skip模式,本文将Skip模式放在PU预测流程的最前面进行。其中,快速Skip算法是通过挖掘残差信息、变换量化系数以及量化参数与Skip模式之间的关系提出的。快速2Nx2N算法是通过分析编码过程中产生的运动矢量残差、变换量化系数中非零数目以及变换单元(Tranform Unit,TU)划分的深度等编码信息与2Nx2N模式之间的关系提出的。实验表明,两个快速模式判决算法都取得了很好的编码效果。同时,由于两个算法针对的是两种不同的PU模式,因此可以将两个算法合并使用。在两个算法合并使用情况下,编码时间在视频质量基本保持不变的基础上在低延时场景下可以减少44%和在随机插入场景下可以减少51%。本文提出的算法通过很好的利用编码过程的已有的编码信息和图像视频自身残差相结合的方法,对帧间CU块划分和PU预测模式的判决提前做出了比较准确的判断,相当程度上加快了HEVC编码器的速度。因此,该算法对HEVC在今后实时系统中的应用具有一定的参考价值。
[Abstract]:With the rapid development of high-definition video applications, produced a huge challenge for video transmission and storage. In this new environment, Joint Collaborative Team on Video Coding (JCT-VC) working group to develop a new video encoding standard High Efficiency Video Coding (HEVC/ H.265.HEVC) to deal with the challenges in order to achieve efficient encoding purpose in the original video encoding standard Advanced Video Coding (AVC/H.264) based on the increase in many more advanced video encoding technology, such as flexible partition, partition the asymmetric motion mode, multi block mode prediction. Compared with H.264, HEVC in the premise to obtain the same video quality can reduce about 50% of the bit stream, but in the encoding complexity of computation is greatly increased. In the wireless mobile terminal equipment general computing power, due to the use of equipment can not support all coding technology, This will greatly reduce the encoding efficiency, thus unable to highlight the HEVC advantage of encoding efficiency. In the video encoding technology in prediction encoding is the calculation part of the largest proportion, of which the interframe prediction encoding is one of the important aspects. Therefore, research on fast encoding algorithms for HEVC frames is one of the most important research contents in today's video encoding in the field. This paper briefly introduces the overall framework and the development of video encoding video encoding standard, and then study the fast encoding of today's HEVC at home and abroad to do a systematic summary. Then, the HEVC codec framework and related key technologies are briefly introduced. Finally, according to the HEVC frame CU (Coding Unit) and PU (Prediction Unit partition) in detail and in-depth study of prediction model for HEVC. In the CU block division process, with the reference of adjacent airspace on L CU (Largest Coding Unit) the maximum depth classification (Depth) for the current LCU delimit a range, thus skipping the PU depth prediction mode decision process, effectively reduce the encoding complexity. In order to further reduce the computational complexity of the CU division, after each depth after PU mode prediction decision, based on the best model the residual information and transform quantization coefficient matrix is proposed ahead of the end of CU division algorithm. For the HEVC PU inter frame prediction mode decision process, this paper presents a fast algorithm for PU model rather than the two biggest - Skip mode and 2Nx2N mode encoding process. In order to predict Skip model, this paper will be the most in front of the Skip mode on the PU in the forecast process. Among them, the fast Skip algorithm is proposed by mining the residual information, the relationship between the transformation and quantization coefficient and the quantization parameter and the model of Skip. Fast 2Nx2N algorithm by analyzing the motion vector residuals generated in the process of encoding, and the number of non-zero transform unit transform quantization coefficient (Tranform Unit TU) proposed the relationship between the depth of encoding information and 2Nx2N modes of the final method. Experimental results show that the two models are found quickly achieved good encoding effect at the same time, because the two algorithms are aimed at two different modes of PU, so it can be used in combination. The two algorithms in the two algorithms combined with case, the encoding time remain unchanged on the basis of the low latency scenarios can be reduced by 44% and can reduce the 51%. algorithm proposed in this paper by good the use of the existing information encoding process of encoding video and image combination in its residual random insertion scene in the video quality, for the inter CU partition and PU prediction mode decision made earlier than A more accurate judgement has speeded up the speed of HEVC encoder to some extent. Therefore, the algorithm has certain reference value for HEVC application in the future real-time system.

【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TN919.81

【参考文献】