The Research Of The Patch Prior Blind Image Deblurring Method Based On The Blur Characteristics

As a medium of information transmission,images are widely used.Due to limited hardware equipment and irresistible external environmental,the quality of the image is degraded.The blur of the image is one of the most important factors causing the degradation of the image quality.Therefore,image deblurring methods are proposed.Blind image deblurring is a process of making the image clear by unknown blur kernel.This method is only based on the observed blurred image to achieve the purpose of deblurring,which is a serious ill-posed problem.Among many methods for solving this,one of the most popular methods is to use the prior knowledge of the image to constrain the solution space of the problem and then restore the clear image.Aiming at the problems existing in the current research methods,thesis studies the characteristics of blurred images and proposes a patch prior image blind deblurring method based on fuzzy characteristics.In this thesis,two blind deblurring schemes are proposed from a special edge image patch to a general blur image patch.Thesis studies the blind deblurring method as follows:Firstly,aiming at the edge-diffusion problem in the process of edge mask acquisition,an improved edge mask extraction method of side window is proposed.The method is based on the characteristics of edge patches,combined with the idea of side window filter and color line model.Based on this,a new image patch blind deblurring method based on edge mask is proposed.In the solution process,an interactive optimization iteration strategy of semi-quadratic splitting is used to optimize the proposed model.This method can enhance the margin of preserving ability and accuracy.These more accurate edges are used for blur kernel estimation,which can improve the accuracy of blur kernel.We conducted verification from two aspects of objective indicators and subjective vision.Through a large number of experiments,it can be found that our method can better solve the blur kernel which is closer to the real blur kernel.The restored image has sharper edges,more detail and better visual effect.Secondly,a new regularization term is proposed for all blurred image patches,not only by the fuzzy characteristics of edge patches,but also by the fuzzy characteristics of all blurred image patches.In the process of observing image blurring,we find that in a local image patch,blurring will reduce the pixel value of the largest pixel,and also increase the pixel value of the smallest pixel.In other words,blurring reduces the maximum difference of pixels within the image patch.This phenomenon is also determined by the inherent characteristics of image blurring.Therefore,an image prior based on local maximum difference value is proposed.This prior can be used to accurately estimate the blur kernel and restore the clarity of the blur image.Thesis mainly proves the validity of the local maximum difference prior from two aspects of mathematical derivation and experimental verification.The proposed method is compared with some successful blind image deblurring methods in recent years.A large number of experiments show that both subjective and objective evaluation perspectives,it is proved that the quality of the restoration results is generally higher and the subjective vision is better.