Research On No-reference Video Quality Assessment Method

With the popularization of computer and maturing of network technology, the network multimedia applications associated with videos have got rapid development. Videos suffer data loss during the process of compression, transmission and storage, and different kinds of distortions are introduced. In order to obtain better video subjective effects, quality evaluation of distortion video is needed, and the parameters of video encoder and transmission channel are adjusted according to the result of video quality assessment. As a result, video quality assessment has become an indispensable part of multimedia system. Human subjective quality evaluation of distortion videos is recognized as the most accurate method. However, the process is so time-consuming for large-scale application. As a result, intelligent analysis of the distortion videos through the mathematical model has become international research hot topics. According to the dependence of the original videos, the objective video quality assessment methods can be divided into three types, which named full-reference, reduce-re fere nce and no-reference. Extra bandwidth is needed to transmit the original video and related information in full-reference and reduce-re ference methods. In contrast, no information about the original video is needed in no-reference method, so it has better flexibility and adaptability, and wider application value. In that context, the research of no-reference video quality assessment is carried out by this thesis.In chapter1, the significance of the topic is explained, and then a summary of current research is made. Finally, the main research topic and the structure of this thesis are introduced.In chapter2, the research of no-reference video quality assessment in the pixel domain is carried out, and a no-reference video quality assessment method based on the distortion estimation is introduced. Firstly, the proposed method calculates the local distortion using the statistical characteristics of the difference between contiguous pixels and the global distortion by measuring the detail loss of the video after Gaussian Filtering, and then the video distortion is estimated by the combination of the local and global distortion. Secondly, the video complexity is calculated through intra and inter prediction. Finally, the objective video quality is obtained utilizing both the video distortion and complexity.In chapter3, the research of no-reference video quality assessment in the compressed domain is carried out, and a no-reference video quality assessment method based on the complexity of video content is proposed. The proposed method calculates the complexity of video content by analyzing the relationship between bit stream, compression ratio and video scene. And then the objective video quality assessment model is formulated by considering the impacts of three key procedures during the compression process; quantization, motion estimation, and bit rate control. Three key factors that stand for their respective characteristics arc then calculated by using the information extracted from compressed bit streams. Finally, the video is cut into several parts with difference scenes by scene change detecting, the video quality of each part is obtained utilizing the model mentioned above, based on which the comprehensive quality of the whole video is calculated.In chapter4, the research of the fundamentals of visual perception characteristics and its application in video quality assessment is carried out, and a no-reference video quality assessment method based on visual perception is introduced. Firstly, the luminance contrast, texture complexity, motion intensity contrast and motion direction consistency sub models are created separately according to the perception of video scene by human visual system (HVS). And then the visual attention model is created combining the temporal and spatial characters based on the sub models mentioned above. Finally, on order to increase the accuracy of the result, the existing no-reference video quality assessment method is improved utilizing the visual attention model by giving different weights to different regions of the video.In chapter5, based on the research of video quality assessment in the pixel and compressed domains, combining with visual perception characters, a two-domain no-reference video quality assessment method based on the visual perception characters is proposed. Firstly, a compressed domain sub video quality assessment model is created based on the encode information are extracted from bit streams, and the video similarity between the distortion and original videos is calculate. Secondly, two distortion artifacts, namely, blockiness and blur are detected in the pixel domian, and then the video distortion is calculated combining with the temporal-spatial visual attention model.Finally, the video quality is given considering both video similarity and distortion.The final chapter concludes the new achievements of the whole research and the prospect of the future research.