Research On Fusion Of Extreme Channels Prior Algorithm For Blind Image Restoration

With the digital technology has being developed,image restoration enjoys numerous applications,such as in surveillance,computational photography and medical imaging.Images are often degraded during the data acquisition process.Often the benefits of improving image quality to the maximum possible extent far outweigh the cost and complexity of the restoration algorithms involved.The problem of restoration of digital images from their degraded measurements plays a central role in a multitude of practically important applications.In image processing,the methods to solve the ill-posed problem include image denoising,image smoothing and deblurring.In order to solve this problem,a lot of research has been done on blind image deblurring.Recently,the alarm about the problem of the extreme channels prior deblurring algorithm on human beings has caused wide public concern.As a result of the extreme channels prior algorithm is easy to produce ringing artifacts and cannot suppress noise in the process of deblurring,therefore,aiming to extreme channels prior deblurring algorithm were researched in this paper.By taking advantage of the total variation model has suppress noise and preserve edges,and the Gaussian filtering model has the advantages of suppressing noise and smoothing images.This paper introduces total variation model and Gaussian filtering model into extreme channels prior,establishes the corresponding optimal model based on Bayesian framework,thus accomplishes dominance complement of the two intellect optimization algorithms.The main research contents of this paper are as follows:1.We propose a blind image restoration based on Gaussian filtering algorithm and dark channel prior mixed regularization.As the average value of dark channel increases with the value of adjacent pixels,the dark channel prior blind image restoration algorithm is easy to produce ringing artifacts and noise.In order to improve the efficiency of dark channel prior algorithm,first of all,Gaussian filtering is incorporated into dark channel prior,and we use L0 regularization reduces the value of the dark channel,which can make smoothing image or remove the ringing effect.Second,the semi-quadratic splitting technique is used to solve the non-convex problem of the model and estimate the latent image.Finally,the Fourier transform is used to calculate the latent image and the blur kernel of the image.Experimental results on blur image data show that the algorithm can protect the texture details and contour structure of the image,prevent the appearance of ringing artifacts,and improve the clarity of the image.2.We propose an effective blind image deblurring model based on total variation of extreme channels prior.When there is no dark pixel in the image,the bright pixel and noise will significantly affect the performance of Gaussian filtering dark channel prior algorithm,and can’t effectively estimate the potential image in the middle.First of all,the proposed method combines the advantages of both the extreme channels prior can effectively solve the problem caused by bright pixels and the total variation algorithm can effectively suppress the image noise.A total variation model is introduced into dark channel and bright channel,and it has been edge preservation of the image and eliminating noise or ringing artifacts.Second,we used alternate minimization of the semi-quadratic splitting technique to solve mathematical models of clear image and blur kernel.Finally,in order to improve the accuracy of the estimate,iterative multi-scale technology updated blur kernel is adopted.On a dataset of blur images,experimental results show that,compared with the most advanced algorithms,visual effect and each evaluation index is significantly improved,and the restored image is closer to the real image.This paper research and improvement the method of extreme channels prior blind image deblurring from two aspects,and carries out experiments on blur image datasets in various scenarios.Compared with the representative methods in recent years,the image restored by the algorithm in this paper is closer to the real image,and its robustness,subjective visual effect and objective evaluation indexes are significantly improved.The simulation results show the feasibility and validity of all the proposed algorithms.