Accelerate Randomized Low Dimensional Manifold Method for Image Reconstruction

In this paper, an image processing method based on low-dimensional manifold model is improved by a structured stochastic method. In the original low-dimensional manifold image processing method, the key step is to solve a large linear system involving the matrix of image patch similarity, but the computation of the similarity matrix and large linear equations solving will bring great computational complexity. In this paper, by using the structural information of the image data, the similarity of image patches is obtained by structured local random computation, so as to quickly calculate the similarity matrix and solve the related linear equations. The experimental results show that the computational efficiency of the proposed method is greatly improved, and results similar to the original method can be obtained.