Image Blur Assessment with Feature Points

Blur is a key factor in the perception of image quality, leading to spread of edges in images. The quantity of feature points extracted from images can represent image shape changes. Compared with sharp images, blurred images tend to contain less feature points, and the reduction of feature points is related to blur. In this paper, we propose a new blind blur assessment metric based on feature points. First, we apply Gaussian blur to the blurred image, producing the re-blurred image. Then feature points from the blurred and re-blurred images are extracted and used to form feature point maps. Next, each feature point map is divided into blocks to compute block-wise quantity map, based on which a feature point similarity map is calculated. Finally, a visual saliency map is employed to conduct the pooling, producing the final blur score. Experimental results on four public databases demonstrate that the predicted blur scores has high correlation with subjective evaluations, and the proposed method outperforms several no-reference image blur metrics, as well as some representative general-purpose blind image quality metrics.