Network video quality assessment based on measuring salient video features

In this paper, we explore the application of saliency information for perceptual quality assessment under different network environment. First we present an experiment to extract the saliency information of video under different network environment. Then we find there exists a relationship between the subjective experience and the video content features. M5' model tree, a data mining method, is good for segmental linearization of single-output multi-output system, was introduced to model the assessment method and its parameters. We combine multiple content features of video using M5' model tree. The final combined model improves the similarity between the subjective and objective assessment.