Objective QoE Prediction for Video Streaming Services: A Novel Full-Reference Methodology

This paper addresses the challenges associated with modeling and evaluation of Quality of Experience (QoE) for content streaming video services. A novel full-reference QoE model is introduced, aiming to predict the user Mean Opinion Score (MOS) for videos streamed from any content streaming platform, although focusing on YouTube. Screen captures of streamed videos were used for an immersive user-experience investigation, avoiding reliance solely on packet analysis. FFmpeg, a widely-used open-source video processing software, and VMAF, an open-source video quality metric by Netflix, are the primary tools used by the proposed approach. The study is carried out in a testbed that allows full control of the radio and network parameters, enabling the emulation of realistic scenarios of mobile networks and establishing actual 5G connections with commercial mobile devices. To gather a diverse range of user opinions and conveniently scale up the dataset, Amazon MTurk is used as a crowdsourcing platform. Data from real users scoring 150 distorted video sequences have been used to train the machine learning algorithm based on QoE model developed in this work. The performance of the proposed model is compared with PEVQ-S, which is a standardized full reference algorithm showing better performance in a comprehensive set of performance metrics.