Image Deblurring Model And Algorithm Research

As an important carrier of information,images play an increasingly important role in various fields.However,in the process of obtaining,processing,transmitting and storing images,the total quality of the image is always degraded due to the influence of the camera and the unpredictable factors of the outside world.In order to meet the needs of high-quality images for image applications and image perception,recovering clear images from degraded images has become a hot topic of research.Image deblurring is an important branch of image restoration,which has great research value and significance.Image deblurring refers to a technique of restoring an original image as much as possible based on the degradation information and image prior information.The total variable regularized image prior model is widely used in image deblurring because of its role in protecting image edge information.However,the model inevitably has a staircase effect,and many scholars have made a lot of improvements to overcome this shortcoming.This thesis draws on the previous experience and combines their advantages to propose a novel image deblurring model.When the proposed model is solved by the alternating direction multiplier method,an adaptive solution method is proposed because of the difficulty of the penalty factor.This thesis mainly works from two aspects of model and algorithm.First of all,this thesis introduces the background and significance of image deblurring,research status at home and abroad,degradation models,image degradation methods and fuzzy types.Secondly,the thesis introduces the total variation regularization model,the overlapping group sparse total variation regularization model and the total variation full norm regularization model.In order to overcome the shortcomings of the total variation regularization model with step effect in image deblurring,this thesis combines the advantages of overlapping groups sparse total variation regularization and total variation full norm regularization,and proposes an overlapping group sparse total variation full norm image deblurring model.The model is solved by a hybrid algorithm of alternating direction multiplier method,majorization-minimization method and threshold method.The experimental results show that the proposed model and the algorithm can recover the clear image better,which not only preserves the edge information of the image,but also greatly reduces the staircase effect.Then,when solving the proposed model for the alternating direction multiplier method,the penalty factor has a great influence on the deblurring problem and is difficult to adjust.Therefore,this thesis adaptively adjusts the penalty factor according to the restored picture when optimizing the model.This method adaptively recovers the best picture while ensuring the calculation speed,and the stability of the algorithm.Finally,the research in this thesis is combined with specific project experiments.Since the blurred traffic signal image is not easy to identify the traffic signal,the recognition rate of the traffic light can be improved by deblurring the blurred image and then identifying the traffic signal.The test results show that the signal recognition can be improved after deblurring,which can improve the accuracy of 5%.