No-reference Video Quality Assessment In Video Conference System

With the rapid development of communication technology and video compressiontechnology, video conference system has also been continually developed and it is widelyused in many fields, such as government, electricity, education, healthcare and financeetc. In the system, the quality of video is one of the two most important indicators, whichinfluence the quality of user experience. And it is also an important factor whichdetermines operation of the normal video conference. At present, the video qualityevaluations for video conference system are still rough and the time-consuming andlaborious subjective evaluation method is mainly used which is not the best choice.In this paper, the factors that influence the video quality will be analyzed. And wewill put forward quality evaluation algorithms based on the analysis of the reconstructedimages for different types of video distortion, such as color cast, fuzzy phenomena andpacket loss to accomplish no-reference video quality evaluation. Color cast is detected byconverting the RGB space of color image to the CIELab space which is closer to thehuman visual characteristics, and analyzing the distribution of chromaticity. For thevideo distortion of fuzzy, a quality evaluation method based on LBT and analyzing theimage details is put forward. Mosaic phenomenon caused by packet loss or bit error isdetected by adopting a method based on the SSIM of adjacent frame and mosaicdetection methods, with which the quality of the video is evaluated.By exploring the video distortions in video conference system, and analyzing thecharacteristics of the different types of distortions, this paper put forwards three videoquality evaluation algorithms based on single image and solves the problems of theexisting algorithms, whose drawbacks are that they are difficult to realize withoutreference video and image. Experimental results show that these video quality evaluationalgorithms put forward in this paper meet the real-time requirements, and the resultsagree well with subjective evaluation results.