Depth Image Denoising Using Graph-Based Sparse Representation

In the field of image processing, image denoising is a very challenging task. The development of graph signal theory provides a new perspective for us to deal with this problem. By converting the image into several image patch groups, this paper turns to study the denoising of the image patch groups by using the graph signal approach, which modulates the image patch group as a set of signal values on the graph and defines the graph Fourier transform to transfer them to the graph frequency domain for denoising. Based on this idea, a graph-based algorithm is proposed for deep image denoising. Experiments show that our method outperforms Block-Matching and 3D (BM3D) filtering, especially in the case of low-level noise.