Defocus blur detection based on multiscale SVD fusion in gradient domain

Abstract

Recently, defocus blur detection has been an extensive study, but it is still full of challenges in the blur estimation without having any prior knowledge of test image such as blur kernel, degree, or camera parameters. Inspired by the observation that the degree of defocus blur depth could be distinguished by different frequencies, a novel blur metric based on Multiscale SVD fusion (M-SVD) is proposed. The blur metric fuses different sub-bands of the selected singular values (SVs) in multiscale image windows, which could drastically reduce the chances of false positives for blur detection and overcome the difficulty that the sharp region is misjudged for a blur region because of its smooth texture. Finally, a blur map is applied on the test image combined with post-processing operation meanshift cluster to segment the blur region. Experimental results demonstrate that the proposed method can detect the defocus blur regions of test images with a satisfactory performance and outperforms the state-of-the-art methods.