No-reference Video Quality Assessment

Digital videos are increasingly finding their way into the day-to-day lives of people due to the rapid proliferation of networked video applications such as video on demand, digital television, video teleconferencing, streaming video over the Internet and so on. Along with that,people have put forward higher requirements on the quality of image and video. Quality control of videos from the capture device to the ultimate human user in these applications is essential in maintaining Quality of Service(QoS) requirements and methods to evaluate the perceptual quality of digital videos form a critical component of video processing and communication systems. Therefore, the research on video quality assessment algorithm has important significance and broad application prospects.At the same time, with continuous study on the evaluation of the video and image, it was found the video image has certain natural scene statistics(NSS).Different distortion and distortion degree of video images lead to different effects on scene statistics. The research work of this paper is based on the two.Based on the two characteristics of HVS and NSS, this paper proposes a reference image quality evaluation method based on deep learning. The algorithm extracts the statistical laws of distorted images and gets the type of distortions using deep learning network. According to the type of distortion, we use the specific method of image quality evaluation to predict the quality of distortion images.Natural video is obtained from the natural image in a certain order, so the natural video has statistical characteristics in the spatial domain. Natural video also has time characteristics. A new no-reference video quality assessment method based on video temporal and spatial characteristics quality evaluation method for natural video is proposed In order to simplify the algorithm, the algorithm first detects the change of the video scene, extracts the video key frame, and obtains the objective quality of the video in the same scene. At the same time, considering the influence of HVS on the human perception, this paper proposes a video quality evaluation based on the region of interest. Firstly, the method detected the video regions of interest and non regions of interest. Then, The block effects and fuzzy effects in the two regions are measured respectively, and the weight of the two different regions is given, and the objective quality of the video is calculated.A large number of experiments on the LIVE image database and VQEG video database, we verify the correctness and the sensitivity of the algorithm proposed in this paper. The test results keep a good consistency with subjective feelings in comparison to the existing methods.