Image Denosing Method Based On The Low Rank Tensor Recovery

The images will always be disturbed by various noises in the process of production or transmission,which can lead to quality degradation,and it will influence the processing of subsequent images,so the images noise removal is an important part of images processing.In recent years,along with the rise of compressed sensing and sparse representation,the theory of low rank restoration has attracted a lot of scholars'interest.So image denoising based on low rank restoration theory becomes a hot topic in this field.On the basis of in-depth study of low rank-sparse theory,an image denoising model based on weighted tensor Schatten p-norm and a low rank-sparse decomposition model based on weighted tensor Schatten p-norm and tensor l₁ norm are established respectively in this paper.The main work of this paper is as follows:1.The limitations of the image denoising model based on the low rank structure of matrix are pointed out.Stacking the similar image blocks to form the third order tensor by the non-local similarity of natural images,and the minimization model based on the weighted tensor Schatten p-norm is established.The optimization problem in this model is solved by using the augmented lagrange multiplier and alternating direction method.And then applied the model to the removal experiment of image Gaussian noise.2.A new model named the a low rank-sparse decomposition model based on the weighted tensor Schatten p-norm and tensor l₁ norm is established,and adaptive center weighted mean filtering method is used to detect the position of random impulse noise and filter.The filtered image is decomposed into a plurality of image sub blocks,which are then stacked into a three-order tensor form.And then the tensor model is established.The augmented lagrange multiplier and the alternating direction method are used to solve each optimization problem in this model.So,the corresponding low rank tensor is obtained.And then expanded it into matrix form to realize denoising.