A Model for Estimating Subjective Evaluation Values of Video Degradation from Viewers’ Physiological Signals

This paper describes that it is possible to estimate a viewer ’ s five-level subjective evaluation of video degradation with an estimation accuracy of 0.90 or better by using physiological data such as blood volume pulses of viewers. To this end, we conducted an experiment to record participants ’ EEG, BVP, gaze, pupil diameter, and subjective evaluation values of video degradation while they watched videos. We then created five different datasets from the data and built estimation models using machine learning based on random forest or neural network. As a result, the coefficient of determination for the physiological data with top importance trained by random forest was 0.997. The results contribute to an objective, continuous, unconscious, and quantitative method for estimating Quality of Experience (QoE) during video viewing.