无参考视频质量评价方法研究
发布时间：2018-12-16 03:57
【摘要】：随着计算机的普及和网络技术的日益成熟,与视频相关的网络多媒体应用得到了迅猛的发展。视频在压缩、传输和存储的过程中会受到数据损失,引入各种失真效应。为了获得更好的视频主观效果,需要评价失真视频的质量,根据评价结果调整编码器和传输信道的相关参数。在目前大多数网络多媒体系统中,视频质量评价已经成为不可或缺的重要组成部分。人类对这些失真视频的主观质量评价被公认为最精确的方法,但是这个过程消耗大量的人力和时间,不适合大规模实际应用。因此,通过设计数学模型对失真视频进行智能化分析,从而计算视频质量的客观质量评价方法,成为当前国际性的研究热点。根据对原始视频的依赖程度,视频客观质量评价方法可以分为全参考型、部分参考型和无参考型三类。由于全参考型和部分参考型评价方法都需要额外的带宽来传输原始视频及相关信息,其实用价值非常有限。相比之下,无参考质量评价方法不需要依赖任何与原始视频相关的信息,直接根据待评价视频的信息计算视频质量,具有更好的灵活性和适应性,以及更广泛的应用价值。本文正是在这样的背景下,展开了对无参考视频质量评价方法的研究。 第一章绪论部分首先阐述了选题的意义,然后综述国内外研究现状并作相应的总结,最后介绍了本课题的主要研究内容和论文结构。 第二章对像素域无参考视频质量评价方法进行研究,提出了一种基于失真度估计的无参考视频质量评价方法。该方法首先统计相邻像素点之间亮度差值的方差,作为视频局部失真度。再对视频进行高斯滤波,计算滤波后的细节损失,得到视频全局失真度。然后,结合这两者估计视频整体的失真度。同时,通过帧内预测和帧间预测计算视频复杂度,反映视频内容特征。最后,综合视频失真度和复杂度得到视频客观质量。 第三章对压缩域无参考视频质量评价方法进行研究,提出了一种基于视频内容复杂度的无参考视频质量评价方法。该方法首先通过分析码率、压缩率与视频场景之间的关系,得到视频内容复杂度。利用宏块模式、量化系数、运动矢量和消耗码流等编码信息计算量化、运动和码流分配影响因子,分别代表视频压缩中量化、运动搜索和码率控制这三个关键环节对压缩后视频质量的影响。然后,在分析这三个影响因子各自对视频质量的反馈的基础上,结合视频内容复杂度,建立基于视频内容复杂度的视频质量评价模型。最后,对整个视频进行场景切换检测,将其分为不同的场景片段,并用此模型评价每个视频场景的客观质量,综合得到整个视频的质量。 第四章对视觉感知特性的基本原理及其在视频质量评价领域中的应用进行研究,提出了一种基于视觉感知特性的无参考视频质量评价方法。首先,根据人类视觉系统(Human Visual System, HVS)对视频场景的感知特性,利用对比度、纹理特征得到空域感知特性,利用运动强度对比度、运动方向得到时域感知特性。然后采用融合时域和空间域特性的方式,建立视觉注意模型。最后用该模型改善已有的无参考视频质量评价方法,根据视觉注意模型对视频的不同部分进行加权计算,提高评价结果的准确度。 第五章在像素域和压缩域无参考视频质量评价的基础上,结合视觉感知特性,提出了一种基于视觉感知特性的双域无参考视频质量评价方法。该方法首先在压缩域提取码流中的编码信息,并利用这些信息建立压缩域视频质量评价子模型,预测失真视频与原始视频的相似度。然后,在像素域检测两种常见的失真效应,块效应和模糊效应的失真程度,并利用第四章提出的时空联合视觉注意模型,对失真效应检测结果进行加权,得到视频失真度。最后,结合视频相似度和失真度给出视频整体的质量评价。 第六章总结了本论文的研究成果和创新点,并提出了进一步研究的方向和任务。
[Abstract]:With the popularization of the computer and the increasing maturity of the network technology, the network multimedia application related to the video has been developed rapidly. The video is lost in the process of compression, transmission, and storage and introduces various distortion effects. In order to obtain a better subjective effect of the video, the quality of the distorted video needs to be evaluated, and the relevant parameters of the encoder and the transmission channel are adjusted according to the evaluation result. In most of the current network multimedia systems, video quality evaluation has become an integral and important component. The subjective quality evaluation of these distorted videos has been recognized as the most accurate method, but this process consumes a lot of manpower and time and is not suitable for large-scale practical applications. Therefore, the objective quality evaluation method of the video quality is calculated by the intelligent analysis of the distorted video by the design mathematical model, which becomes the current international research hotspot. according to the degree of dependence on the original video, the video objective quality evaluation method can be divided into a full-reference type, a partial reference type and a non-reference type. Since the full-reference and partial-reference-type evaluation methods require additional bandwidth to transmit the original video and related information, the value is very limited. In contrast, the non-reference quality evaluation method does not need to rely on any information related to the original video, and directly calculates the video quality according to the information of the video to be evaluated, and has better flexibility and adaptability and wider application value. It is in this background that the research on the method of no-reference video quality evaluation is carried out. The first chapter introduces the significance of the topic, then summarizes the domestic and foreign research status and the corresponding summary, and finally introduces the main research contents and the thesis of the subject. In the second chapter, the non-reference video quality evaluation method for pixel domain is studied, and a non-reference video quality based on distortion estimation is proposed. The method comprises the following steps of: firstly, counting the variance of the brightness difference value among the adjacent pixel points, local distortion is obtained, the video is subjected to Gaussian filtering, the filtered detail loss is calculated, and the video is obtained Global distortion. Then, combine the two to estimate the video the distortion of the body is calculated by the intra-frame prediction and inter-frame prediction, and the video complexity is reflected, and finally, the comprehensive video distortion degree and the complexity are obtained. The third chapter studies the method of non-reference video quality evaluation in the compressed domain, and puts forward a non-reference based on the complexity of video content. The method comprises the following steps of: firstly, analyzing the relationship between a code rate, a compression rate and a video scene, the coding information such as the macro block mode, the quantization coefficient, the motion vector and the consumption code stream is used for calculating the quantization, the motion and the code stream allocation influence factors, and then, on the basis of analyzing the feedback of the three influence factors on the video quality, combining the video content complexity and establishing the video content complexity, and finally, carrying out scene switching detection on the whole video, dividing the whole video into different scene segments, and evaluating the objective quality of each video scene by using the model, The fourth chapter studies the basic principle of visual perception and its application in the field of video quality evaluation. The method for evaluating the non-reference video quality comprises the following steps of: firstly, according to the perception characteristics of a human visual system (HVS) on a video scene, in the dynamic direction, the time-domain sensing characteristics are obtained, and finally, using the model to improve the existing non-reference video quality evaluation method, and carrying out weight calculation on different parts of the video according to the visual attention model, In chapter 5, based on the non-reference video quality evaluation of the pixel domain and the compressed domain, a visual perception characteristic is proposed based on the visual perception characteristic. The method comprises the following steps of: extracting the encoding information in a code stream in a compressed domain, and establishing a compressed domain video quality evaluation sub-model by using the information, wherein the method The similarity of the distorted video to the original video is measured. Then, the distortion degree of two common distortion effects, block effect and fuzzy effect is detected in the pixel domain, and the time-space combined visual attention model proposed in the fourth chapter is used to detect the distortion effect. the result is weighted to obtain the video distortion, and finally, the video similarity and the video similarity are combined, In chapter 6, the research results and the innovation of this paper are summarized in the sixth chapter.
【学位授予单位】：浙江大学
【学位级别】：博士
【学位授予年份】：2012
【分类号】：TP391.41
本文编号：2381784
[Abstract]:With the popularization of the computer and the increasing maturity of the network technology, the network multimedia application related to the video has been developed rapidly. The video is lost in the process of compression, transmission, and storage and introduces various distortion effects. In order to obtain a better subjective effect of the video, the quality of the distorted video needs to be evaluated, and the relevant parameters of the encoder and the transmission channel are adjusted according to the evaluation result. In most of the current network multimedia systems, video quality evaluation has become an integral and important component. The subjective quality evaluation of these distorted videos has been recognized as the most accurate method, but this process consumes a lot of manpower and time and is not suitable for large-scale practical applications. Therefore, the objective quality evaluation method of the video quality is calculated by the intelligent analysis of the distorted video by the design mathematical model, which becomes the current international research hotspot. according to the degree of dependence on the original video, the video objective quality evaluation method can be divided into a full-reference type, a partial reference type and a non-reference type. Since the full-reference and partial-reference-type evaluation methods require additional bandwidth to transmit the original video and related information, the value is very limited. In contrast, the non-reference quality evaluation method does not need to rely on any information related to the original video, and directly calculates the video quality according to the information of the video to be evaluated, and has better flexibility and adaptability and wider application value. It is in this background that the research on the method of no-reference video quality evaluation is carried out. The first chapter introduces the significance of the topic, then summarizes the domestic and foreign research status and the corresponding summary, and finally introduces the main research contents and the thesis of the subject. In the second chapter, the non-reference video quality evaluation method for pixel domain is studied, and a non-reference video quality based on distortion estimation is proposed. The method comprises the following steps of: firstly, counting the variance of the brightness difference value among the adjacent pixel points, local distortion is obtained, the video is subjected to Gaussian filtering, the filtered detail loss is calculated, and the video is obtained Global distortion. Then, combine the two to estimate the video the distortion of the body is calculated by the intra-frame prediction and inter-frame prediction, and the video complexity is reflected, and finally, the comprehensive video distortion degree and the complexity are obtained. The third chapter studies the method of non-reference video quality evaluation in the compressed domain, and puts forward a non-reference based on the complexity of video content. The method comprises the following steps of: firstly, analyzing the relationship between a code rate, a compression rate and a video scene, the coding information such as the macro block mode, the quantization coefficient, the motion vector and the consumption code stream is used for calculating the quantization, the motion and the code stream allocation influence factors, and then, on the basis of analyzing the feedback of the three influence factors on the video quality, combining the video content complexity and establishing the video content complexity, and finally, carrying out scene switching detection on the whole video, dividing the whole video into different scene segments, and evaluating the objective quality of each video scene by using the model, The fourth chapter studies the basic principle of visual perception and its application in the field of video quality evaluation. The method for evaluating the non-reference video quality comprises the following steps of: firstly, according to the perception characteristics of a human visual system (HVS) on a video scene, in the dynamic direction, the time-domain sensing characteristics are obtained, and finally, using the model to improve the existing non-reference video quality evaluation method, and carrying out weight calculation on different parts of the video according to the visual attention model, In chapter 5, based on the non-reference video quality evaluation of the pixel domain and the compressed domain, a visual perception characteristic is proposed based on the visual perception characteristic. The method comprises the following steps of: extracting the encoding information in a code stream in a compressed domain, and establishing a compressed domain video quality evaluation sub-model by using the information, wherein the method The similarity of the distorted video to the original video is measured. Then, the distortion degree of two common distortion effects, block effect and fuzzy effect is detected in the pixel domain, and the time-space combined visual attention model proposed in the fourth chapter is used to detect the distortion effect. the result is weighted to obtain the video distortion, and finally, the video similarity and the video similarity are combined, In chapter 6, the research results and the innovation of this paper are summarized in the sixth chapter.
【学位授予单位】：浙江大学
【学位级别】：博士
【学位授予年份】：2012
【分类号】：TP391.41
【引证文献】
相关期刊论文 前1条
1 周传兴;涂巧玲;张杰;张晗;;基于视觉感知的网络视频质量评价方法研究[J];微型机与应用;2017年11期
相关硕士学位论文 前6条
1 崔明明;视频底层特征选取及其与观众评价的相关分析[D];东北电力大学;2016年
2 贾琳;基于DSP的视频质量监测系统设计[D];黑龙江大学;2015年
3 卢培磊;无参考视频平滑度评价方法的研究[D];武汉工程大学;2015年
4 袁媛;基于显著性区域的视频质量评价[D];北京邮电大学;2015年
5 马瑞泽;压缩立体视频质量客观评价方法研究[D];天津大学;2014年
6 刘立冬;基于H.264压缩域的视频水印算法研究[D];浙江大学;2013年
，本文编号：2381784
本文链接：https://www.wllwen.com/kejilunwen/xinxigongchenglunwen/2381784.html
