No-reference hybrid video quality assessment based on partial least squares regression

Abstract

Constructing objective assessment model on visual quality is challenging, since it associates closely with many factors in human visual perception, as well as both source coding and transmission. In this paper, a no reference hybrid model for video quality assessment is proposed by employing Partial Least Squares Regression (PLSR). The hybrid model combines both bitstream-based features and network-based features, taking into video quality dependence on visual content and network conditions account. We have conducted subjective tests to validate its quality assessment accuracy. Simulation results in Network Simulator version 2 (NS-2) verify the performance advantage of the hybrid model for SD (standard definition) and HD (high definition) video sequences. Proposed model could be used as a quality monitoring tool for assessing the real-time video quality during wireless transmission. Simulation results show that the devised model can improve the quality prediction against a full-reference metric, and achieves a high correlation (0.955) between quality prediction and the actual perceived quality judged by subjective assessment.