High Order Image Priors Learning For Image Deblurring

Digital image processing has received widespread attention in recent years, with the popularity of all kinds of electronic products in our daily life. As an important component of the image processing field, image deblurring has been hotspot internationally. Image blurring is usually caused by the shaking of people’s hands or scene movement, that is, the so called motion blur.In many deblurring methods, one of the most common priors is that the gradient magnitudes of natural image obey heavy-tailed distributions, generally as a regularization item in the stage of image restoration. However, the traditional deblurring methods only take the first-order prior of natural images into consideration, without considering the higher order prior. Meanwhile, details would be missed in the restored image due to the inaccuracy of the traditional approximation model.Aiming at these problem, we have extensively studied the FoE model firstly, through which we can get eight3x3filters which contain high order priors of natural images; second, we use these filters for guiding the image restoration, which changes the traditional regularization term based on heavy-tailed distribution, and propose a restoration model based on high order priors; finally, we adopt the IRLS method to solve our model.Experiments show that our method based on FoE (Field of Experts) model of image restoration proposed in this paper has achieved good results, which can keep the image details very well in image restoration stage at the same time. Moreover, our image restoration methods is efficient which performs well even for large blurs.