Subjective Media Quality Recovery from Noisy Raw Opinion Scores: A Non-Parametric Perspective
Résumé
Recent parametric approaches to recover subjective media quality from noisy opinion scores involve assumptions that are challenged by numerous complex factors influencing viewer behavior in subjective tests. These approaches commonly involve modeling scoring behavior using parametric distributions, with parameters estimated through methods like maximum likelihood or Bayesian techniques. This paper proposes instead a nonparametric perspective to address the media quality recovery problem, without making any a priori assumption on the subjects' scoring behavior. We propose and motivate an approach to measure the reliability of each single opinion score of a subject. The proposed measure of reliability is then leveraged to formulate an Entropy-based Subjective Quality Recovery (ESQR) algorithm. Simulations and experiments on real datasets show that the proposed ESQR algorithm compares favorably to several state-of-the-art approaches in terms of robustness to noise across various testing conditions. The code to run ESQR and reproduce the results in this paper is made freely available at: http://media.polito.it/ESQR.
Origine | Fichiers produits par l'(les) auteur(s) |
---|