Évaluation de la qualité perceptuelle des signaux multimédias.

Abstract

Quantifying image quality without reference is still a challenging problem, especially when different distortions affect the observed image. One no-reference metric is not able to assess the different distortions presented in the images. In this work, we have focused on the assessment based on a fusion scheme of multiple distortion measures. These metrics are built in two stages. First, a set of relevant image quality assessment metrics is used to evaluate the perceptual quality. Then, a support vector regression (SVR)-based fusion strategy is adopted to derive the overall index of image quality. For the 2D metric, we have used the particle swarm optimization to select the appropriate metrics. While for the 3D metric, we have incorporated the binocular vision in the assessment process. To evaluate the performance of the proposed metrics, different image databases have been used. The obtained results demonstrate clearly that the proposed approaches outperform the stateof-the-art NR-IQA methods. Furthermore, the proposed approaches are flexible and could be extended to other distortions.