网络视频业务平均意见分统计研究及在流分类中的应用

发布时间：2018-08-06 12:41

【摘要】：随着网络视频传输技术和无线网络技术的发展,人们对于视频业务的需求变得越来越大。如何根据所提供的服务来准确预测用户的体验质量是迫切需要解决的问题。一方面,网络资源的有限性及信道存在的差错率等因素,使得准确地预测网络环境下用户观看视频的体验质量有着较大挑战。而传统的服务质量(Quality of Service,QoS)是强调技术层面的客观评价指标,不能直接体现用户对视频业务质量的真实感受。另一方面,为了更好的控制和管理网络流量以及保障网络视频的服务质量,保障用户的视频体验质量,找到高效、简单的统计特征组合对网络视频业务进行有效的识别与分类一直是研究的重点。本文主要研究工作的内容如下:首先,本文详细介绍了主观质量评价方法和客观质量评价方法的理论及视频体验质量(quality of experience,QoE)的定义,阐述了部分现有的视频业务用户体验质量评估方法。然后,详细介绍了ITU-T P.1201标准提出的客观质量评价模型和矢量量化算法。根据ITU-T P.1201用户感知的媒体流QoE预测模型,通过使用矢量量化算法分析了三种分辨率视频业务的视频码率、平均播放中断时间、播放中断次数、初始缓冲时间等QoS参数对用户体验质量的影响。针对该预测模型综合考虑了单独的音频评分、单独的视频评分、音视频评分、播放中断的评分、整体的媒体段评分,本文研究了三种分辨率视频业务关于平均意见分(Mean Opinion Score,MOS)的概率密度函数(Probability Density Function,PDF)。最后,针对上述分析的三种质量视频业务关于MOS值的PDF,本文提出了一种基于三种质量视频业务MOS值PDF均值特征的流识别与分类方法。该方法研究了使用视频MOS值相关特征进行网络视频业务的识别与分类。通过综合考虑CON-GR特征选择方法的最优特征子集、视频业务MOS值PDF均值的特征,采用支持向量机(Support Vector Machine,SVM)分类器实现了标清、高清和超清三类视频的细粒度分类。实验结果表明,该方法可以获得较高的视频分类准确率。
[Abstract]:With the development of network video transmission technology and wireless network technology, the demand for video service becomes more and more. How to accurately predict the user's experience quality based on the services provided is an urgent problem to be solved. On the one hand, the limitation of network resources and the error rate of channel make it a great challenge to accurately predict the experience quality of users watching video in network environment. The traditional quality of service (Quality of) is an objective evaluation index in the technical level, which can not directly reflect the users' true feelings about the quality of video services. On the other hand, in order to better control and manage the network traffic and ensure the quality of service of the network video, ensure the quality of the user's video experience, find out the high efficiency, Simple statistical feature combination has been the focus of research on the effective recognition and classification of network video services. The main contents of this paper are as follows: firstly, this paper introduces the theory of subjective quality evaluation and objective quality evaluation and the definition of video experience quality (quality of. This paper describes some existing methods of quality evaluation of user experience in video services. Then, the objective quality evaluation model and vector quantization algorithm proposed by ITU-T P.1201 are introduced in detail. According to the QoE prediction model of media stream perceived by ITU-T P.1201, this paper analyzes the video bitrate, average playback interrupt time and the number of playback interruptions of three kinds of resolution video services by using vector quantization algorithm. The effect of QoS parameters such as initial buffer time on user experience quality. For the prediction model, a comprehensive consideration is given to the individual audio score, the individual video score, the audio and video score, the playback interruption score, and the overall media segment score. In this paper, we study the probability density function (Probability Density) of three kinds of resolution video services about average opinion score (Mean Opinion ScoreMOS). Finally, this paper proposes a stream recognition and classification method based on the PDF mean feature of MOS value for three kinds of quality video services analyzed above. This method studies the recognition and classification of network video services using video MOS value correlation features. By considering the optimal feature subset of CON-GR feature selection method and the feature of PDF mean of MOS value in video traffic, the support vector machine (SVM) classifier is used to realize the fine-grained classification of three kinds of video: clear, high-definition and super-definition. Experimental results show that this method can achieve a higher accuracy of video classification.
【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】：TP393.09

【相似文献】