No-Reference Quality Assessment of Deblurred Images Based on Natural Scene Statistics

Abstract

Blurring is one of the most common distortions in digital images. In the past decade, extensive image deblurring algorithms have been proposed to restore a latent clean image from its blurred version. However, very little work has been dedicated to the quality assessment of deblurred images, which may hinder further development of more advanced deblurring techniques. Motivated by this, this paper presents a no-reference quality metric for defocus deblured images based on Natural Scene Statistics (NSS). Two categories of NSS features are extracted in both the spatial and frequency domains to account for both the global and local aspects of distortions in deblurred images. Specifically, the spatial domain NSS features are used to characterize the global naturalness, and the frequency domain NSS features are used to portray the local structural distortions. All features are combined to train a support vector regression model for quality prediction of defocus deblurred images. The performance of the proposed metric is evaluated in a subjectively rated defocus deblurred image database. The experimental results demonstrate the advantages of the proposed metric over the relevant state-of-the-arts. As an application, the proposed metric is further used for benchmarking deblurring algorithms and very encouraging results are achieved.