基于流媒体的视频质量评价模型

发布时间：2018-06-25 07:24

本文选题：视频质量评价 + 模糊效应　；参考：《中山大学》2014年硕士论文

【摘要】：随着流媒体技术的发展与普及，网络视频成为了主流的多媒体载体。网络视频在流媒体传输过程中需要经过压缩、传输、解压等过程，这类操作会造成视频质量损伤。网络视频的质量决定着流媒体的服务质量，因此越来越多的流媒体运营商需要一种直观准确的基于流媒体的视频质量评价模型以实时获取网络视频质量值，从而掌握当前流媒体的性能，并能选择合适的流媒体服务器以及对流媒体技术进行提升。因此，对流媒体视频的质量评价的研究有着十分重要的意义。由于视频质量的无参考评价法无需参考原始视频，且适用于批量处理及具有较好的时效性。本文主要对视频质量的无参考评价法进行深入研究，并将本文使用的评价法运用于流媒体系统中。首先，提出基于线性预测误差的模糊效应度量法。使用最小二乘法求出图像分块的线性组合，使用该线性组合去计算像素预测值与原始值的差值作为模糊效应度量的特征值。其次，，改进基于周期性块检测的方块效应度量法和基于小波变换的噪声效应度量法。对块效应的度量增加三个限制条件以更好的区别真实纹理与块纹理，从而提高块效应度量的可靠度。对噪声效应的度量采用三种带有浮动阈值的小波处理，提高了噪声效应度量的有效性。然后，使用机器学习法整合三类失真效应，给出视频质量综合评价方法。结合机器学习法对带有主观评价值的数据库进行训练，并把通过机器学习法总结出不同效应所引起的不同视觉感受的特性运用于三类失真效应的整合上，最终给出视频质量综合评价方法。最后，实现基于流媒体的视频质量评价模型及谷歌视频搜索插件。结合流媒体技术、网络视频爬虫技术、视频关键帧提取技术以及视频质量综合评价法，实现基于流媒体的视频质量评价模型。为证明模型的可用性以及时效性，将结合谷歌应用扩展，实现基于谷歌视频搜索的视频质量评价插件，根据用户的视频搜索结果，实时调用评价模型的数据库并将结果通过网页反馈给用户。
[Abstract]:With the development and popularization of streaming media technology, network video has become the mainstream multimedia carrier. In the process of streaming media transmission, network video needs to be compressed, transmitted, decompressed and so on. This kind of operation will cause video quality damage. The quality of network video determines the quality of service of streaming media, so more and more streaming media operators need an intuitive and accurate video quality evaluation model based on streaming media in order to obtain the quality of network video in real time. In order to master the current streaming media performance, and can choose the appropriate streaming media server and streaming media technology to improve. Therefore, the research on the quality evaluation of streaming media video is of great significance. Because the non-reference evaluation method of video quality does not need to refer to the original video, it is suitable for batch processing and has good timeliness. In this paper, the non-reference evaluation method of video quality is studied deeply, and the evaluation method used in this paper is applied to streaming media system. Firstly, a fuzzy effect measurement method based on linear prediction error is proposed. The linear combination of image blocks is obtained by using the least square method, and the difference between the pixel prediction value and the original value is calculated as the eigenvalue of the fuzzy effect measurement. Secondly, the block effect measurement method based on periodic block detection and the noise effect measurement method based on wavelet transform are improved. In order to improve the reliability of block effect measurement, three restrictions are added to distinguish real texture from block texture. Three kinds of wavelet processing with floating threshold are used to measure noise effect, which improves the effectiveness of noise effect measurement. Then, the machine learning method is used to integrate three kinds of distortion effects, and a comprehensive evaluation method of video quality is presented. Combined with machine learning method, the database with subjective evaluation value is trained, and the characteristics of different visual feelings caused by different effects are summed up by machine learning method and applied to the integration of three kinds of distortion effects. Finally, a comprehensive evaluation method of video quality is presented. Finally, the video quality evaluation model based on streaming media and Google video search plug-in are implemented. The video quality evaluation model based on streaming media is realized by combining streaming media technology, network video crawler technology, video key frame extraction technology and video quality comprehensive evaluation method. In order to prove the usability and timeliness of the model, a video quality evaluation plug-in based on Google Video search will be implemented in combination with the Google application extension, according to the user's video search results, The database of the evaluation model is called in real time and the result is fed back to the user through the web page.
【学位授予单位】：中山大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN919.8

【参考文献】