Various algorithms have been proposed for image inpainting,among which sparse representations of image inpaintin g via an overcomplete dictionary have recently produced promising results.Since the images are all non-negative,the traditional dictionary learning methods imposed simply with non-negativity may become inapplicable.In this paper,a non-negative dictionary learning method for image inpainting is presented,and the proposed method enforces sparseness by penalizing the L0norm of the coefficient matrix in non-negative factorization objective function.The method consists of two steps:a sparse coding and a dictionary update stag.A nonnegative version of orthogonal matching pursuit (OMP) is used for s parse coding,and the dictionary update resembles standard K-SVD.With learned dictionary,Smoothed L0(SL0) algori thm is used for image reconstruction.Numerical experiments were conducted,and the results d emonstrate that the proposed inpainting method outperforms conventional K-SVD image inpainting algorithm and several other algorithms for comprasion,and it achieves state-of-art inpainting performance in terms of b oth peak signal to noise ratio (PSNR) and structural similarity index measurement(S SIM).What’s more,its computation c onsumption is fairly low.