网络丢包的无参考视频质量评估方法研究

发布时间：2018-03-02 11:41

本文选题：无参考视频质量评估　切入点：网络视频　出处：《西安电子科技大学》2014年博士论文　论文类型：学位论文

【摘要】：IP(Internet Protocol)网络和移动通信的迅猛发展,使得视频信息在IP网络和移动通信网络中的处理和传输受到越来越多的关注。如何对视频质量进行准确有效的评估,已成为在视频压缩、处理以及传输等研究领域迫切需要解决的问题之一网络丢包是导致视频质量下降的主要原因之一,论文针对网络丢包的无参考视频质量评估进行了深入的研究,主要研究工作着眼于提高客观频质量评估的实时性和有效性。分别从H.264/AVC编码新特性、丢帧类型、丢包率、包丢失集中度和人眼视觉容忍特性对视频质量的影响等方面进行了深入的研究。主要研究内容包括： 1. H.264/AVC网络视频的丢包失真评估。针对由H.264/AVC所使用的新编解码技术引起信道误码、空时域误码传播、空时域误码掩盖及去方块滤波对视频质量的影响,给出了一种计算网络视频丢包失真MSE(Mean Square Error)的无参考视频质量评估算法,该算法充分考虑了空时域误码传播、误码掩盖算法和编码器特性对视频质量的影响,并且计算复杂度低,可以进行实时网络视频质量监控。实验结果表明,使用该算法能够较准确地反映视频质量MSE。并结合视频内容空时域特性,对全参考视频质量评估方法SSIM(Structural Similarity Metric)进行改进,给出一种基于SSIM的半参考视频质量评估算法。实验结果表明,使用该算法得到的客观视频质量能够较准确地反映视频的主观质量。 2.考虑丢帧类型的无参考视频质量评估模型。针对网络视频质量评估需要满足实时性和低资源开销等要求,给出一种无参考网络视频质量评估模型。该模型考虑了丢失不同帧类型、视频内容和丢包影响单帧时人眼视觉容忍特性对视频质量的影响,且不需要对视频完全解码,具有资源开销小、实时性好的特点,适合对实时传输的视频流进行质量评估。实验结果表明,使用该视频质量评估模型测得的视频丢包失真质量评分与其主观质量评分具有很好的一致性,相比于国际标准G.1070中的丢包视频质量评价模型,该模型与主观质量评价的相关性平均提高了6.96%。 3.考虑包丢失特性的无参考网络视频质量评估模型。针对相同包丢失率不同包丢失集中度对视频质量影响不同的问题,提出一种考虑包丢失特性的无参考网络视频质量评估模型。该模型依据码流信息在压缩域计算视频的时域复杂度,以及包丢失率对视频质量的影响因子；为了解决包丢失集中度对视频质量的影响,提出在压缩域计算包丢失集中度的方法,进而提出由于丢包引起的视频质量下降的视频质量评估模型,因此适用于具有版权保护的视频序列质量评估,并且资源开销小。实验结果表明,使用该模型测得的视频包丢失质量评分与其主观质量评分有很好的一致性,相比于国际标准G.1070中的丢包视频质量评价模型,该模型与主观质量评价的相关性平均提高了6.18% 4.考虑人眼视觉容忍特性的无参考网络视频质量评估模型。视频通信的最终信宿是人,对视频质量的评价取决于人眼的主观感受,因此合理的视频质量评估方法必须符合人眼的视觉特性,给出一种考虑人眼视觉特性的无参考视频质量评估模型。首先对网络丢包、误码传播以及包丢失在视频帧内的位置对视频帧质量的影响进行了详细的研究；然后分析了网络丢包影响多帧时的人眼视觉容忍性对视频序列质量的影响；最后通过对视频序列时域失真的研究进而得到整个视频序列的客观质量。该模型充分考虑人眼视觉系统的空时域掩盖效应及容忍性、视频内容、包丢失位置、视频序列的时域失真对视频质量的影响,并且该模型只对接收的码流进行简单的包头分析即可得到所需信息,因此具有资源开销小、实时性好的特点,适合对实时传输的视频流进行质量监控。实验结果表明,使用该视频质量评估模型测得的视频丢包失真质量评分与其主观质量评分有很好的一致性。
[Abstract]:IP (Internet Protocol) the rapid development of Internet and mobile communication, the video information processing and transmission in IP network and mobile communication network has attracted more and more attention. How to accurately and effectively assess the video quality, has become a research field in video compression, processing and transmission of the urgent need to address the problem of packet loss is one of the main causes of the decline in the quality of video, the network loss no reference video quality assessment conducted in-depth research, the main research focuses on the real-time and effective improve the objective quality assessment. The frequency characteristics of H.264/AVC encoding, frame type, packet loss rate, packet loss and concentration of human vision have influences on the video quality and other aspects of the in-depth research. The main research contents include:
1. packet loss H.264/AVC network video distortion evaluation. According to the new decoding technique used by H.264/AVC caused by channel error, spatial and temporal error propagation, spatial and temporal error concealment and deblocking filter effects on the video quality, a calculating network video packet loss MSE (Mean Square Error) distortion no reference video quality assessment algorithm. This algorithm considers the spatial and temporal error propagation, error concealment algorithm and the influence on the quality of video encoder characteristics, and low computational complexity, real-time network video quality monitoring. The experimental results show that this algorithm can accurately reflect the quality of the video and video content with MSE. space-time domain characteristics, method of SSIM to evaluate the reference video quality (Structural Similarity Metric) was improved, given an evaluation algorithm of semi reference video quality based on SSIM. The experimental results show that The objective video quality obtained by this algorithm can accurately reflect the subjective quality of the video.
2. consider no reference video quality assessment model for frame type. In view of the network video quality assessment needs to meet the real-time and low resource cost requirements, presents a no reference video quality assessment model. The model considers the loss of different frame types, video content and loss effect of single frame of human visual tolerance affect the characteristics of video the quality, and does not need to fully decode the video resources, with small overhead, good real-time performance, suitable for real-time transmission of video stream quality assessment. The experimental results show that the use of the evaluation model of the measured packet loss of video quality score and subjective quality score distortion is well consistent with the video quality, packet loss video quality assessment model compared to the international standard of G.1070, the correlation model and subjective quality evaluation by an average of 6.96%.
3. consider the non reference video quality assessment model of packet loss characteristics. According to the loss rate of different concentration of packet loss on the video quality of different problems with the same package, a new no reference video quality assessment model considering the packet loss characteristics. The model based on information stream video in the compressed domain time domain computational complexity. The impact factor and packet loss on the video quality; in order to solve the packet loss concentration effects on the video quality, put forward to calculate the concentration of the lost packet in the compressed domain, and then put forward the evaluation model of packet loss caused by the decline due to the video quality of video quality, so it is suitable for video quality assessment with copyright protection, and resources low cost. The experimental results show that using this model the measured packet loss quality score and subjective quality scores are in good agreement, compared to the international The quality evaluation model of packet loss video in standard G.1070, the correlation between the model and the subjective quality evaluation is increased by 6.18%.
4. consider the non reference video quality assessment model of human visual tolerance characteristics. The final destination of video communication is one of the video quality evaluation on subjective feelings in the human eye, so the video quality assessment method is reasonable and must be consistent with the human visual characteristics, gives a kind of considering human visual characteristics no reference video quality assessment model. Firstly, the network packet loss, packet loss and error propagation effects in the video frame position within the frame of video quality are studied in detail; and then analyzes the human visual effects of multi frame network loss tolerance effect on the quality of the video sequence; finally, through the research on objective quality distortion of video sequence to obtain the entire video domain sequence. This model fully considers the space-time domain of human visual system and masking tolerance, video content, packet loss, video sequence The time domain distortion effect on the video quality, the required information and get the model only on the received bit stream by simple analysis to Baotou, so it has low cost of resource, good real-time performance, suitable for real-time transmission of video stream quality monitoring. Experimental results show that the video packet loss using the video quality assessment model the measured distortion quality score and subjective quality scores are in good agreement.

【学位授予单位】：西安电子科技大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TN919.81

【参考文献】