Research On Subjective Video Quality Assessment

Video quality assessment plays important roles in video compression, video processing, and video communications where compression and transmission always lead to quality degradation. Since subjective video quality assessment is the intuitive reflection of video quality in observers'perception and improving the subjective quality is the final goal of video communications and video services, subjective assessment serves as the basis of all other assessment methods and attracts extensive research interests.Subjective quality assessment experiments require observers to score the video sequences according to the given rating scale. The choice of rating scale directly influences the observers'correct expression of their perception of video quality, and the consistency of the rating results. Choosing a proper rating scale is therefore a key step of subjective quality assessment. Firstly, the rating scale should reflect the sensitivity of the human visual system. It is therefore necessary to find a proper rating scale which is consistency with human perception. Furthermore, due to the subjective understanding of videos could be different for different observers, as well as other issues such as the lack of concentration in the process of subjective evaluation, there would be outliers in the subjective test scores. To ensure the reliability of subjective evaluation results, post-processing of the data is needed to eliminate the outliers. This thesis is devoted to the design of the rating scale and appropriate data post-processing. The major contributions of the related research work include:1. Build experimental environment and develop test platform for subjective video quality assessment following the requirements of international standards.2. Based on the theory of visual perception and signal detection, perform research on the relationship of human visual sensitivity and different rating scales. Compare the impact of different rating scales on the confidence interval of subjective test scores, analyze and design the rating scale which is consistent with human perception.3. Analyze the importance and disadvantages of the current post-processing technique, and propose a new algorithm considering the confidence interval which eliminates the outliners with higher accuracy.