An Ensemble Learning-Based No Reference Qoe Model For User Generated Contents

Duc V. Nguyen, Huyen T. T. Tran, Truong Cong Thang

Abstract: The explosive growth of user-generated content (UGC) videos demands new Quality of Experience model for this new type of content. In this paper, we propose a novel No Reference QoE model for UGC videos. The proposed model takes into account not only bitstream-level and packet-level information, but also encoding parameters. Features are extracted from input videos at frame level, then combined by a pooling layer. In addition, ensemble learning is utilized to improve overall prediction performance. Experimental results show that our model can predict MOS values of UGC videos with very high accuracy. Also, the proposed model is simple, and so can be used for real-time QoE monitoring.