Image De-noising Algorithm Based On Matrix Regression

During the photography and transmission process,the quality of the image is inevitably affected by the device or the transmission medium.We call it a noisy image,which will seriously affect the visual effect and the further processing of the image.Therefore,in recent decades,image de-noising has become one of the important research topics in the field of image processing.The image de-noising method based on matrix regression is a supervised learning method based on preserving the intrinsic internal structure information of the image and taking the image matrix as the basic unit.This paper mainly contains the following contents:(1)For the single image super-resolution problem,Tang et al.proposed an image reconstruction method based on matrix value regression.Inspired by this,a single-side linear operator de-noising method based on matrix regression is proposed for image de-noising.The global linear operator obtained by training directly acts on the noise image,which is simple,easy to implement,and time-consuming.Experiment shows that the single-side linear operator method based on matrix regression is suitable for multiple noise(Gaussian,salt and pepper,periodic noise)images,and has obvious removal effect on mixed noise.(2)The single-side linear operator de-noising algorithm based on matrix regression has limited de-noising effect on mixed periodic noise.The reason is that the global linear operator only processes the raw or columns of the image in the de-noising process.To this end,we assume that the mixed periodic noise image matrix is the result of linear transformation of row and column vectors of the original image matrix,and then derive the global bilateral linear operator acting on both sides of the matrix,and propose a bilateral linear operator algorithm based on matrix regression.In addition,in order to prevent over-fitting,the bilateral linear operators need to be properly constrained.Here we use l2 norm regularization term to achieve this.The experimental results show that the algorithm is very effective in removing periodic noise,even for the case of high noise intensity.(3)Non-local mean filter algorithm deals with noise by combining local and global information of image.Because the method based on matrix regression essentially uses global information,the third task of this paper is to use the idea of non-local mean and its improved algorithm for reference,to model noise image using the local information,to train global linear operator through matrix regression,and then to propose an adaptive weighted mean de-noising method based on matrix regression.The algorithm combines local and global information of the image,for the test image with same type of noise and similar noise intensity,its de-noising effect is better than the adaptive median filter.