Support Vector Regression Based Video Quality Prediction

Abstract

To measure the quality of experience (QoE) of a video, the current approaches of objective quality metrics development focus on how to design a video quality model, which considers the effects of the extracted features and models the Human Visual System (HVS). However, video quality metrics which try to model the HVS confronts a fact that HVS is too complicated and not well understood to model. In this paper, instead of modeling the objective quality metrics with some functions, we proposed to build a video quality metrics using the support vector machines (SVMs) supervised learning. With the proposed SVM based video quality prediction, it allows a much better approximation to the NTIA-VQM and MOS values, compared to the previous G.1070-based video quality prediction. We further investigated how to choose the certain features which can be efficiently used as SVM input variables.