Research On Image Denoising Algorithm Based On Low-rank And Sparse Decomposition

Image denoising is a significant research filed in image processing,and it greatly affects other advanced image processing applications such as super-resolution,image matching,object recognition and so on.In order to meet the needs of following work and improve the quality of image,image denoising has become a very important work in the field of image preprocessing.The development of low rank and sparse decomposition theory opens up a new direction for image denoising.In this paper,the mainstream denoising algorithm based on the low rank matrix sparse decomposition is introduced,and some improved algorithms are proposed to improve the effect of the algorithm.Firstly,a method of image denoising based on the method noise is proposed,in view of the fact that the existing denoising algorithms not only remove the noise but also remove part of the details,which make the image blurring.This paper analyzes the method noise and establishes a model to extract image information submerged in the method noise.Then the extracted image information is fed back to the denoised image to conduct the next denoising step.Experimental results show that the proposed method is superior to other state-of-the-art methods in terms of objective and subjective quality performance.And it is suitable for all the denoising algorithms using the iterative theory.Compared with three existing mainstream denoising algorithms,can achieve good results.Secondly,in the weighted kernel-norm minimization algorithm,the number of matching blocks is selected according to the experience,which is artificial setting rather than considering adaptive image characteristics.In the process of gaining weight,the weighted kernel-norm minimization algorithm only consider the influence of noise intensity,which leads the weights too rough.This paper proposes a weighted kernel-norm minimization algorithm based on gradient.First,calculate the gradient of a noisy image,and divides the image into smooth area and detail area,then select corresponding similar sets for different regions.At the same time,the concept of image gradient is imported to the process of weight setting to improve the reasonability of weight,and then improve the effect of denoising.Finally,aiming at the phenomenon of that the setting of terminating conditions is not reasonable in most of the existing image denoising algorithms based on iterative theory,this paper proposes an iterative termination strategy based on noise estimation.Estimating the intensity of noise is the first step in the process of denoising.If the iteration process cannot reduce the noise,what we should do is to stop the iteration.Because in this case the iterative operation is invalid,and even introduce more noise.Using the proposed iterative stopping strategy based on noise estimation,there is no need for prior artificial presupposition.At the same time,the method realize the image adaptive.On the basis of ensuring the denoising effect,the operation efficiency is improved at the same time.