基于结构相似性的客观视频质量评价

发布时间：2018-04-28 03:19

本文选题：视频质量评价 + 结构相似度　；参考：《新疆大学》2017年硕士论文

【摘要】：随着移动互联网的快速发展,视频通信逐步成为人们日常生活中传递和交流信息的重要载体,尤其现在社交媒体爆发式增长,微博、微信中存在大量的视频通信,用户在通信中对视频图像的质量要求也相应提高。然而视频图像在采集、编码、传输过程中都有可能产生失真,这时需要对呈现在用户面前的视频质量进行测试,以便优化系统,提供更好的视觉体验。从用户角度出发,主观视频质量评价更真实,但此方法要求较多的人力物力,无法应用到实际的评价系统中,效率较低,而客观的视频质量评价方法,在评价过程中不能完全符合人的主观视觉感受,研究更加符合用户真实体验的客观视频质量评价成为重要的课题。人眼的视觉系统可以提取视觉场景中的结构信息,结构相似度方法突破性的利用人眼的这一特性来评价视频图像质量,这一方法主要是通过场景中结构信息的变化来度量视频图像的失真。本文仔细研究了结构相似度理论以及具体应用过程中的方法,并在此基础上进行了两方面的改进。第一,针对结构相似度算法评估严重失真的模糊图像时效果不理想,本文提出采用直方图表现单帧图像的结构特征,单帧图像的模糊程度用直方图的集中度表征,再与运动估计结合来评价视频质量。第二,针对结构相似度算法在评价过程中没有考虑感兴趣的问题,本文把视频图像分为感兴趣区域和不感兴趣区域,由于在感兴趣区域边缘结构比较丰富,所以在此区域采用边缘结构相似度的方法,在不感兴趣区域采用结构相似度方法,最后将两个评价结果进行加权处理。这样可以区分出视频图像中的重要信息,具有很好的评价效果。实验结果表明,基于改进的结构信息和运动估计的方法在评价严重失真的视频时比单尺度的结构相似度方法有更好的效果。基于视觉感兴趣区的评价方法比其他算法更加符合人眼特性,主客观评价的一致性更好。
[Abstract]:With the rapid development of mobile Internet, video communication has gradually become an important carrier of information transmission and exchange in people's daily life, especially the explosive growth of social media, Weibo, WeChat, there are a large number of video communications, The quality requirement of video image is also improved in communication. However, in the process of acquisition, coding and transmission of video images, distortion may occur. In order to optimize the system and provide a better visual experience, it is necessary to test the video quality in front of users. From the user's point of view, subjective video quality evaluation is more realistic, but this method requires more manpower and material resources, and can not be applied to the actual evaluation system. It is inefficient and objective video quality evaluation method. In the process of evaluation, the objective video quality evaluation which is more in line with the user's real experience has become an important subject. The human visual system can extract the structure information from the visual scene. The structural similarity method uses this characteristic of the human eye to evaluate the video image quality. This method mainly measures the distortion of video image by changing the structure information in the scene. In this paper, the theory of structural similarity and the method of its application are studied carefully, and two improvements are made on this basis. First, in view of the unsatisfactory effect of structural similarity algorithm in evaluating seriously distorted fuzzy images, this paper proposes a histogram to represent the structural features of a single frame image, and the fuzzy degree of a single frame image is characterized by the histogram concentration degree. And then combined with motion estimation to evaluate the video quality. Secondly, because the structure similarity algorithm does not consider the problem of interest in the evaluation process, this paper divides the video image into the region of interest and the region of interest, because of the rich edge structure in the region of interest. So the method of edge structure similarity is adopted in this region, and the structure similarity method is used in the region of no interest. Finally, the two evaluation results are weighted. In this way, the important information in the video image can be distinguished, and the evaluation effect is very good. The experimental results show that the improved structural information and motion estimation method are more effective than the single-scale structural similarity method in evaluating seriously distorted video. The evaluation method based on the region of visual interest is more consistent with the human eye than other algorithms, and the subjective and objective evaluation is more consistent.
【学位授予单位】：新疆大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN919.8

【相似文献】