Quality estimation based on interest points through hierarchical saliency maps

Abstract

Quality of Experience (QoE) is a widely used notion nowadays because the end-user has been re-integrated in the quality loop. Subjective experiments are tedious and time consuming but to date they are the main way to have the human judgment. An important effort has been put on the development of metric estimating the QoE. So, in this paper, we propose a new image quality metric based on two concepts: the interest points and the objects saliency on color images. This metric is constructed by taking advantage of the variability of the interest points impact function of the image impairment and also the hierarchical attention of a human observer. This combination helps in giving more importance to the variability of the most salient regions and reducing the influence of regions having less visual importance.The results show a high correlation between the metric scores and the human judgment and a better quality range than well-known metrics.

Investigation and modeling of visual fatigue caused by S3D content using eye-tracking

“…It is closely connected with the attention, hence with the semantic information provided by the scene. In our work, saliency maps based on interest points were chosen [50]. To the best of our knowledge, there is no information about scene classification according to its semantic load.…”

A statistical study of the correlation between interest points and gaze points

“…The aim is to determine whether the interest points can be used to predict salient informations on an image like the HVS does. This can help for several applications like quality assessment, 10 simplified saliency maps construction, 11 . .…”