Research On Depth Image Denoising Algorithm Of Joint Graph Model

With the rapid development of the Internet,the depth images are increasingly becoming the hot research topic nowadays because they reflect the three-dimensional(3D)scene and can be applied in various fields of computer vision.But the depth images obtained from depth camera usually contain stains such as noise,which greatly impairs the performance of depth related applications.It is particularly important to design a set of denoising algorithm for depth images.In recent years,the methods exploring the image intrinsic property have gradually becoming the popular denoising algorithm.In this paper,our proposed algorithm is based on the graph model of exploring image intrinsic self-similarity property.This paper firstly introduces the sparse property of the image and proposes the united graph model and sparse representation denoising method.In the proposed united method by exploring the self-similarity and sparsity property,a vector-based dual graph model is constructed by treating each vector of image block as a node.Meanwhile,the classical separated Bregman iteration algorithm(SBI)of convex optimization is used to solve the united equation,which greatly improves the efficiency and speed of the solution.In order to expand the denoising method which combines with the graph model further,this paper proposes a united graph model and nuclear norm denoising method by introducing the low rank property of images.In the proposed method by exploring self-similarity and low rank property,a learning-based graph model is proposed by using the manifold learning method.Alternating direction multiplier method(ADMM)algorithm is used to solve the united equation for convex optimization.Meanwhile,a newly fast threshold algorithm is proposed for solving sub-problems.Finally,an overall regularized iteration is proposed to optimize the denoising results further.The final experimental results show that two denoising methods which combines with the graph model acquire clearer edges in the subjective visual effect compared with other state-of-the-art denoising methods such as BM3D,NLM and WNNM.Meanwhile,the proposed methods achieves higher peak signal-to-noise ratio(PSNR)and structural similarity(SSIM)than other method.