Quality Assessment of Blur Images Via Saliency and Multiscale Features

摘要: Blur is the most common distortion in images, so it significant to evaluate quality of blur images. Considering view saliency and multiscale characteristics human vision, this paper proposes a fast efficient assessment method. Firstly, spatial sharpness maps are extracted, map used as weight map. At same time, images downsampled multiscales texture naturalness image on each scale measured. Then all features train support vector regression model. Finally, model employed Experimental results conducted four databases demonstrate that comparing with existing methods general metrics, presented method has obvious superiority performance computation speed.