视频质量客观评价算法研究

发布时间：2018-05-07 09:51

本文选题：视频质量评价 + 全参考　；参考：《天津大学》2016年硕士论文

【摘要】：由于网络视频应用的快速发展,数字视频逐渐走近我们的日常生活,如视频点播、数字电视、视频会议、网络流媒体视频和视频监控等。从前端视频获取设备到用户终端,为了保证视频应用服务质量和用户体验,我们需要从视频处理和传输等环节检测视频图像质量。由于视频影像最终受体往往是人眼,任何一个有生活经验的普通人都能较为准确地评价视频的清晰度、连贯性、颜色的鲜艳程度、图像饱和度等。因此人眼主观评价是最准确的视频质量评价途径。然而,对于如今巨大的视频数据量,主观视频质量评价已经力不从心。因此采用数学方法和计算机程序进行自动化视频质量评价的客观评价方法,已经成为视频质量评价领域的研究热点。客观视频质量评价方法根据算法对原始视频的依赖程度分为全参考、部分参考和无参考评价算法。本文研究重点是全参考与无参考视频质量评价算法,主要工作分为三个部分:1.在对视频空域与时域视觉感知特性研究的基础上,引入三维梯度相似度改进了全参考视频质量评价算法——基于时空域梯度相似度的视频质量评价算法,并将其与当前国际上的全参考视频质量评价算法进行比较,实验结果表明本算法具有较好的评价性能和较低算法复杂度;2.对国际上当前最好的通用型无参考图像质量评价算法进行综述。首先介绍了每一种算法的特征提取和质量评价原理,然后在LIVE主观质量评价数据库上对几种算法进行了仿真实验,最后定量地和定性地评价了这几种算法的性能,分析了各自算法的优势与不足;3.调研了视频监控系统中存在的常见视频空域、时域失真类型,并将通用型无参考图像质量评价算法应用到视频监控系统中,最后将客观评价结果与主观评价结果进行了对比。
[Abstract]:With the rapid development of network video applications, digital video gradually approaches our daily life, such as video-on-demand, digital television, video conferencing, network streaming video and video surveillance. From the front-end video acquisition device to the user terminal, in order to ensure the quality of service and user experience of video application, we need to detect the video image quality from video processing and transmission. Because the ultimate receptor of video image is usually human eye, any ordinary person with life experience can accurately evaluate the definition, consistency, bright color, image saturation and so on. Therefore, subjective evaluation of human eyes is the most accurate way to evaluate video quality. However, subjective video quality evaluation has been inadequate for the huge amount of video data. Therefore, the objective evaluation method of automatic video quality evaluation using mathematical methods and computer programs has become a research hotspot in the field of video quality evaluation. Objective video quality evaluation method is divided into full reference, partial reference and non-reference evaluation algorithm according to the degree of dependence of the algorithm on the original video. This paper focuses on the full reference and no reference video quality evaluation algorithm, the main work is divided into three parts: 1. Based on the research of visual perception characteristics in spatial domain and time domain, 3D gradient similarity is introduced to improve the full reference video quality evaluation algorithm, which is based on temporal and spatial gradient similarity. The experimental results show that this algorithm has better evaluation performance and lower complexity. This paper reviews the best universal image quality evaluation algorithms in the world. Firstly, the principle of feature extraction and quality evaluation of each algorithm is introduced, then several algorithms are simulated on the LIVE subjective quality evaluation database. Finally, the performance of these algorithms is evaluated quantitatively and qualitatively. The advantages and disadvantages of each algorithm are analyzed. The common video spatial and temporal distortion types in video surveillance system are investigated, and the general non-reference image quality evaluation algorithm is applied to the video surveillance system. Finally, the objective evaluation results are compared with the subjective evaluation results.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】