Research On Blind Deblurring Algorithm For Single Motion Blurred Image

Image is the basis of human vision and one of the most common information carriers in human social activities,and it is indispensable in daily life.However,due to the influence of external environment and internal irresistible factors,digital image degradation occurs in the acquisition process,which brings troubles to the subsequent research,analysis and application.Therefore,blind deblurring of degraded images has become one of the hot fields in modern image processing.Since the 20 th century,the research on blur image deblurring algorithm has been continuous,which can be divided into two categories: deconvolution deblurring algorithm based on image prior and deblurring algorithm based on deep learning.In this paper,the motion blurred images are taken as the research object,and the image deblurring algorithms are studied by combining the prior knowledge such as L0 regularization,low-rank prior,salient structure and high-order total variation.The main research results are as follows:An adaptive iterative blind image deblurring algorithm combining regularization and low-rank prior is proposed to solve obvious ringing phenomenon and inaccurate blur kernel estimation in traditional deconvolution algorithms.First,the sparseness of L0 regularized prior is used to estimate the intermediate clear images and remove ringing effect effectively.At the same time,low-rank prior is introduced to suppress the noise interference in the process of latent image restoration to improve the accuracy of blur kernel estimation.Then,the number of iterations is adjusted by comparing the similarity between the blur kernel models,which solves the problem of the fixed number of iterations at each scale under the pyramid layer and effectively reduces the calculation cost.Finally,the mathematical model is an inverse process with high discomfort and non-unique solutions.The algorithm model uses augmented Lagrange method,semi-quadratic splitting method and least square method to alternate optimization solutions,and then uses non-blind deblurring method to get the final clear image.The experimental results of the proposed algorithm are compared with other classical algorithms,and combined with the image quality evaluation index data to verify the effectiveness of the proposed blind deblurring algorithm model.An image blind deblurring algorithm based on gradient cepstrum and multi-regularization constraints is proposed to solve the problems of ringing and ladder phenomenon,missing edge details and high computational cost in the traditional image blind deblurring method based on single sparse regularization prior.This algorithm extracts the salient pixel information of the middle potential image by the image salient structure prior,and uses the high-order and low-order total variation regularization constraint to suppress the ladder and ringing phenomenon that are prone to occur in the process of latent image restoration and protect the image edge pixel information.Gradient cepstrum strategy is used to initialize the number of iterations in the blur and adjust the multi-scale to effectively reduce the computational cost.The research of this algorithm is divided into two parts: one is to establish a new algorithm model for optimization and solution,the other is to verify the superiority of the proposed algorithm by comparing the experimental results.Through experiments on synthetic data and natural blur scenes,it is found that the proposed algorithm can effectively estimate the blur kernel and suppress the effect of ring and ladder at the same time.Moreover,it has rich features of edges and details,and the running speed is also improved.Both subjective vision and objective image quality evaluation data show the superiority of the proposed algorithm.There are 31 pictures,12 tables,and 57 references.