DE LA PERCEPTION LOCALE DES DISTORSIONS DE CODAGE A L'APPRECIATION GLOBALE DE LA QUALITE VISUELLE DES IMAGES ET VIDEOS. APPORT DE L'ATTENTION VISUELLE DANS LE JUGEMENT DE QUALITE

摘要

This study deals with the local evaluation of perceptual distortions, the overall visual quality assessment and the influence of visual attention in visual quality assessment. To locally evaluate distortions in images, we have simplified an existing human visual system model using wavelet transform and we have proposed an improved visual masking model that takes into account both semi-local masking and contrast masking. From these models, we have designed and tested several image quality metrics. Regarding videos, we have developed a new method to locally evaluate the spatio-temporal distortions. This method is based on a short-term temporal pooling of spatial distortions which simulates the evaluation of distortions through some selection mechanisms of visual attention. A video quality metric based on this method has been designed and validated. It is based on a longterm temporal pooling incorporating perceptual saturation and asymmetric behavior. In order to study visual attention in subjective and objective visual quality assessment, eye-tracking experiments on images and videos have been conducted both in free task and quality task. From collected data we have studied the visual attention deployed in the different configurations. The results have confirmed, among others, the influence of the quality task on deployment of visual attention. The impact of visual attention in the construction of the quality judgment has also been studied using the real saliency information. Results show that, both with images and videos, a simple linear weighting of distortions by the visual attention does not clearly improve performances of objective quality metrics.