Research On Algorithm Of Motion Blurred Image Restoration Based On Mixed Regularization

As an important carrier of receiving outside information image plays an increasingly important role in people's working life.Especially with the rapid development of artificial intelligence,face recognition,intelligent home and automatic driving,it is very important to obtain a clear image of high quality.However,for a variety of reasons,such as the rapid moving of the objects or camera and other hand-held photographic equipment jitter,the images will generate motion blur,which brings great inconvenience to the image of the user.Further,when the blur kernel is unknown,the image blurring appears more difficult.Therefore,the blind deblur of image has important application significance in the present society.The main work of this thesis is to propose a blind motion blurred image restoration algorithm based on mixed regularization,which is as follows:1.Aiming at the question of inaccurate PSF estimation when blind motion image was deblurred,an image restoration algorithm based on the hyper Laplace prior and the higher-order total variation regularization is proposed according to the sparsity of the heavy-tailed distribution of natural images.The hyper Laplace prior can reflect the heavy-tailed distribution characteristics of natural priors most effectively and recover the edge information of the image effectively.Besides the high-order total variation can effectively remove the staircase effect and ringing effect that are easy to appear in the smooth region.The combination can get closer PSF to the real one and effectively restore the global information of the image,especially for the greater degree of blur image.The subjective and objective comparison experiments prove that the algorithm can effectively recover the image.2.Aiming at the problem of weak edge protection for the restored image for normalized sparse priors,the paper adds the higher-order and lower-order full-variance regularization terms on the basis of normalized sparse priori regularization according to the sparse theory of signal and the regularization theory.The lower order TV regularization model can protect the edge sparsity of the natural image,and the High-order TV regularization model can restrain the ladder and the ringing effect of the image's non edge part.The experimental results show that the algorithm can effectively recover the blurred image with more complex edge information and protect the edge details of the image with suppressing the ringing effect simultaneously.It is more robust to the visual quality and objective quality evaluation by comparing with other methods for motion blur image blind restoration.3.Aiming at the multi-parameter and non-convex characteristics of this algorithm,a method of combining split Bregman iteration algorithm and GST algorithm is proposed.Because there are many parameters in the algorithm and it is a multi-parameter optimization problem,a split Bregman iterative algorithm is used to solve the multi-parameters.The multi-parameter problem is converted into multiple sub-problems to be solved separately,which reduces the difficulty of the solution.To solve the non-convex hyper Laplace(l_p)problem,the paper adopts the GST algorithm,which can easily solve the optimal solution for any P(0<P<1)value.The combination of the two methods can greatly reduce the solution time and improve the optimization efficiency..