Research On Blind Motion Debluring

With the rapid development of computer technology and the rising popularity of smart mobilephones,digital images exist everywhere in people's lives and work.During the process of image acquisition,generally,certain relative motion happening between imaging devices and shooting scenes,will lead to image blurring,greatly influencing people's work and life.In recent years,the problems of blind motion image deblurring constantly received extensive attention of various fields at home and abroad,and in public safety,traffic safety monitoring,the aerospace,medical image,reconnaissance,military and many other important fields,the blind image deblurring technology are all needed.Therefore,this subject is full of significant research value and profound realistic significance.The main work of blind motion image deblurring lies in the estimation of both the sharp image and blur kernel from the observed blurred image.This dissertation focuses on blind motion image deblurring with respect to global motion blur and mainly carries out the following work.1.Considering that the estimated latent image still contains relatively serious blur and noise,and the direct use of the image edge information fails to correctly guide estimating blur kernel,we extract salient structures from the latent image,and use the salient structures will be used to estimate the blur kernel.In view of this,a blind motion image deblurring algorithm,based on the NGM?Non-dimensional Gaussianity Measure?sparsity prior and edge constraint,is proposed.Restore the latent image under the Non-dimension Gauss Measure regularization constraints first and then extract the salient edge with the help of T-smooth technology.Finally,using the extracted salient structural information,obtain the comparatively accurate blur kernel through a variational Dirichlet blur kernel estimating method.During the non-blind deblurring stage,considering the advantages and disadvantages of total variational and super-Laplacian regularization,we reduce the ringing effect of the restored image and retain more image details at the same time by averaging the images recovered from the above two methods.2.Considering about the sparsity of the latent image gradient,we introduce an adaptive l₀-norm regularization to constrain the latent image gradient,recover the salient structure of the latent image,and guide to estimate the blur kernel according to the salient structure information.Considering the sparsity,continuity and normalized feature of the blur kernel,we construct a new kernel blur kernel constraints with the l₀-norm and l₂-norm,and constrain the solution space of the blur kernel through the regularizations,which produce a blind motion image deblurring algorithm based on adaptive regularization.In order to verify the effectiveness of the proposed algorithms,a large number of experiments are carried out on the Levin standard test data and the actual color blurry images respectively.The experiments show that the proposed algorithms in this paper achieve a better result of blind deblurring.