A Study Of Weighted Regularization Term Based Image Denoising

Image processing technology is a powerful tool in robot vision,face recognition,safety monitoring,artificial intelligence,medical imaging and other fields.The overall performance of image processing system depends on the quality of the test image,but the image is inevitably affected by noise in the process of acquisition and transmission.The purpose of image denoising is to reconstruct the image from the image data polluted by noise.It can extract image data better and improve the quality of degraded image.Therefore,image denoising is the basic problem and important process of many image processing systems.In recent decades,image denoising has attracted much attention.Researchers found that the method based on variational partial differential equation has significant efficiency in the field of image denoising.In this paper,a new denoising model is proposed to improve the total variation regularization term,because the classical ROF model is easy to blur the image edge with the number of iterations.In addition,an effective algorithm is designed to solve the new model.In order to evaluate the denoising effect of the new model proposed in this paper,several groups of contrast experiments are designed.The experimental results show that the new model can effectively protect the edge details of the image while denoising.The main work of this paper includes the following three aspects:1.A new additive noise removal model is proposed.In this model,we derive a weighted regularization term,which can be regarded as the balance between Total Variation regularization term and a nonconvex regularization term with strong edge preserving performance.On the one hand,the weighted regularization term inherits the advantages of the Total Variation regularization term and can effectively remove the noise in a homogeneous region of an image;on the other hand,it inherits the edge preserving performance of the nonconvex regularization term and can protect the edge geometry of image.2.An effective algorithm is proposed to solve the new model.In this paper,the iteratively reweighting algorithm is used to transform the proposed new model into two subproblems,and then the chambolle projection algorithm coupled with the alternating direction multiplier method(ADMM)is used to solve the related subproblem.In addition,the effectiveness of the new model is verified by a number of comparative experiments.The numerical results show that the weighted regularization term model proposed in this paper can produce better denoising effect compared with the ROF model and the nonconvex regularization term model.3.Through some reasonable assumptions and rigorous theoretical derivation,we prove the existence and uniqueness of the minimal solution of the weighted regularization term model,and further prove the convergence of the suquence generated by the alternating direction multiplier method(ADMM)which is used to solve one of the subproblems.