Predicting Personal Preferences in Subjective Video Quality Assessment

In this paper, we study the problem of predicting the visual quality of a specific test sample (e.g. a video clip) experienced by a specific user, based on the ratings by other users for the same sample and the same user for other samples. A simple linear model and algorithm is presented, where the characteristics of each test sample are represented by a set of parameters, and the individual preferences are represented by weights for the parameters. According to the validation experiment performed on public visual quality databases annotated with raw individual scores, the proposed model can predict the scores by individuals more accurately than the average score for the respective sample computed from the scores given by other individuals. In many cases, the proposed algorithm also outperforms more generic Parametric Probabilistic Matrix Factorization (PPMF) technique developed for collaborative filtering in recommendation systems.