A content-adaptive video quality assessment method for online media service

Abstract

Video quality assessment is an important issue for Internet Content Providers (ICPs) to improve their service. Some research has been done on objective video quality assessment, but real-time evaluation is still a difficult task. This paper discusses a real-time content-adaptive evaluation method to evaluate the Quality of Experiment (QoE) for online media services. The method is named as Motion Degree of Video Content (MDVC) which is defined to make the video content measureable and computable. It analyzes the relationship between the information entropy gain and the frame size of I, P and B frames, and then evaluates the motion degree of a video clip. Then with the help of a multimedia service simulation platform, a QoE evaluation model is established to map network QoS to QoE. The model is adjusted by MDVC so as to fit different video content dynamically. In particular, to ensure that the evaluation model fits the real-world conditions, PlanetLab is adopted to monitor the real QoS on Internet. Finally we compare the content-adaptive QoE evaluation model with the actual MOS values to verify the feasibility and fitness of the model. Results show that the correlation coefficient reaches 0.91.