节点文献

基于视觉优化的视频编码相关技术研究

Researches on the Video Coding Related Techniques Based on Visual Optimization

【作者】 骆政屹

【导师】 郑世宝; 宋利;

【作者基本信息】 上海交通大学 , 信号与信息处理, 2013, 博士

【摘要】 随着信息技术的发展,视频在现代生活中的应用越来越广泛。但由于视频的数据量非常庞大,实际应用中往往需要进行视频的压缩编码处理,由此也促进了与视频编码相关的各项技术的研究与应用。由于人眼经常是视频的最终接收者,在视频编码相关技术中考虑视觉特性往往具有重要的意义。目前,虽然已有许多考虑视觉特性的视频编码相关的研究成果,但随着新近的视频编码标准和传输保护方法的推出,与之相适应的基于视觉特性的视频编码相关技术仍有一定的研究价值。本文即围绕基于视觉优化的视频编码相关技术展开讨论。视频编码相关的研究主要集中于编码、传输、解码三个方面。本文也分别从这三个方面选择研究内容。在编码阶段考虑视觉特性一般是施行主观特性编码。本文首先提出一种针对H.264/AVC标准的基于视觉特性的视频编码方法。我们首先利用经典DCT域的恰可察觉失真(Just Noticeable Distortion,JND)模型估计出视觉的敏感性分布。然后,我们把相应的JND阈值转换至H.264/AVC变换域中,并根据转换后的阈值自适应地抑制预测残差的变换系数。此外,我们还从等效失真的角度修正了用于率失真优化的拉格朗日(Lagrange)乘子,以使其适应带有系数抑制的编码。所提出的方法与H.264/AVC标准完全兼容。实验结果表明,在保持相似视觉质量的前提下,该方法在量化参数(Quantization Parameter,QP)低于28时能比JM14.2参考软件明显节省码率,在多数情况下也比Chen的方法有更好的码率节省效果。由于一般的信道往往无法避免传输差错的影响,因此视频传输时通常需要相应的差错保护措施。本文接着针对最近开发出的喷泉(Fountain)信道码--Raptor码,研究基于Raptor码的非均等差错保护(Unequal Error Protection,UEP)方法。与许多已有的方法不同,我们采用标准的Raptor码以获得更好的实用性。我们利用Raptor码在分组长度较长时的良好特性以及视频包的优先级,分别提出了复杂度不同的基本型和增强型非均等差错保护方法。并且也进一步给出了所提方法的优化配置算法。实验结果表明,在丢包环境下,我们的方法能比GRIP方法更好地提升视频的客观和主观质量。在解码阶段提升视觉效果经常采用误码掩盖的方法。本文最后根据人眼往往更关注图像中感兴趣区(Region Of Interest,ROI)的特性,提出一种基于H.264/AVC标准的改进感兴趣区误码掩盖的方法。在编码端,根据背景宏块的量化后系数,我们把感兴趣区宏块的运动矢量信息自适应地嵌入到背景中。考虑到背景中只能嵌入有限量的信息,我们进一步提出根据一种预定义的误码传播度量来确定感兴趣区宏块的嵌入优先级。提出的方法与H.264/AVC标准完全兼容,并且可以与许多其它的误码掩盖方法结合使用。实验结果表明,该方法不会对编码效率产生大的影响,但能比DMVE算法和改进的BMA算法更好地改善感兴趣区的误码掩盖效果。

【Abstract】 With the development of information technologies, video applications have gainedincreasing popularity in modern life. But in view of the huge size of video data, videocoding is often necessary in practice for compression purposes. And researches andapplications on video coding related techniques are also promoted correspondingly.Since human eyes often act as the ultimate receivers of video, it’s often desiredto take into account the visual characteristics in video coding related techniques. Nowalthough there have been many research achievements for video coding related tech-niques considering visual characteristics, with the introduction of new video codingstandards and transmission protection methods, corresponding video coding relatedtechniques based on visual characteristics still deserve further researches. And thisthesis just puts its emphasis on the researches on the video coding related techniquesbased on visual optimization.Researches on video coding related techniques focus mainly on encoding, trans-mission and decoding. And this thesis also selects the research topics from these threeaspects.Usually visual characteristics are taken into account at the encoding stage byperceptual coding. In this thesis, firstly a method of perceptual coding based on theH.264/AVC standard is proposed. First, the distribution of visual sensitivity is estimat-ed through a Just Noticeable Distortion (JND) model in the classic DCT domain. Thenthe corresponding JND thresholds are translated to those of the H.264/AVC transformdomain, which are used to suppress the transform coefficients of prediction residualsadaptively. Besides, the Lagrange multiplier for rate distortion optimization is alsoadapted to coding with coefficient suppression in terms of equivalent distortion. Theproposed method is fully compatible with the H.264/AVC standard. Experimental results show that at similar visual quality, the method can obtain obvious bit savingcompared with the JM14.2reference software when the quantization parameter (QP)is less than28, and can obtain more bit saving than Chen’s method in most cases.As common channels usually suffer from transmission errors inevitably, suitableerror resilient methods are often necessary during video transmission. In this thesis,secondly in view of the recently developed fountain channel codes—Raptor codes,Unequal Error Protection (UEP) methods based on Raptor codes are explored. Dif-ferent from many existing methods, standard Raptor codes are adopted here for betterpracticality. By making use of the good characteristics of Raptor codes at large blocklength and the priorities of video packets, we propose respectively a baseline and anenhanced UEP method, which have different complexities. Besides, we also give anefficient algorithm to optimize the configuration of the proposed methods. Experimen-tal results show that both the objective and the subjective quality of video can be betterimproved by our methods than the GRIP method in case of packet losses.Error concealment is often utilized at the decoding stage to improve the visualquality. In this thesis, thirdly according to the property that human eyes usually paymore attention to region of interest (ROI) in images, an improved error concealmentmethod of ROI based on the H.264/AVC standard is proposed. At the encoder side,the motion vector information of ROI macroblocks is embedded in the backgroundbased on original quantized coefficients of background macroblocks. Considering thelimited embedding capacity of the background, we further propose to determine theembedding priorities of ROI macroblocks based on a predefined metric of error prop-agation. The method is fully compatible with the H.264/AVC standard, and can be ap-plied jointly with many other error concealment methods. Experimental results showthat the method can improve the error concealment of ROI better than the DMVEalgorithm and the improved BMA algorithm without much loss of coding efficiency.

节点文献中: 

本文链接的文献网络图示:

本文的引文网络