Image Deblurring Based On Adaptive Super Laplacian Priors

As an important medium for production and life,has a wide range of applications.In the process of image acquisition,the image is blurred because of camera shake,object movement,and the blur can give people a lot of inconvenience.Image blind deblurring is an important branch of image preprocessing.In recent years,it has received extensive attention in the field of scientific research and application.The current,blind deblurring of images is mostly based on Bayesian theory.There are mainly two types of image deblurring methods.One is based on the variation Bayes,one is with the maximum a posteriori estimate.Variational Bayesian-based methods are more robust but computationally inefficient.MAP-based methods generally requires a regular term to constrain the solution space.The existing approaches usually rely on carefully designed fixed regularizers and handcrafted parameter tuning to obtain satisfactory estimation of the blur kernel.In addition,many regularizers exhibit the smoothing capability,but fail to enhance salient edges.In order to better fit the prior distribution of the data,we proposed a novel learning-based blind deconvolution method.We learn a Multi-Scale Shrinkage Fields model(MSSF).At each scale,we obtain the nonlinear functions and parameters through the data-driven way.This algorithm completely avoids design a priori items and manual tuning.The learned model can be applied directly without additional training.Training-based methods are often limited by the choice of training set.In order to solve this problem,we proposes a more flexible adaptive prior method.This method can use different a priori when processing one image.The purpose is to better select the salient edges and enhance the salient edges in the iterative solution process.As the iterative solution process proceeds,the method progressively estimates the kernel and the latent image.We proved the convergence of this method.Many experiments proved that the methods of this paper is effective and efficient.