Text Image Deblurring via Intensity Extremums Prior

Abstract

A novel and effective blind text image deblurring approach which takes advantage of the intensity extremums prior is proposed in the work. Our method is inspired by the phenomenon that the black and white pixels in blurred images are less than the corresponding clear images, especially for text images. And the intensity extremums prior is proved mathematically in this paper. To deblur text images by the intensity extremums prior, an effective optimization algorithm which utilizes a half-quadratic splitting strategy is exploited. Besides the experiments on the document images, the introduced algorithm is also examined on complex text images which contain cluttered background regions, and the results manifest that our approach has outstanding performance against some state-of-the-art image deblurring methods.