Graph-Regularized NMF with Prior Knowledge for Image Inpainting
The image inpainting problem can be converted to the matrix completion. A classical matrix completion method is based on matrix factorization. The product of two low-rank matrices fills in the missing regions. In this paper, we propose a novel matrix factorization framework to recover the images. Before decomposing the original matrix, approximation matrix as the prior knowledge is constructed to estimate the values of missing pixels. The estimation of the missing pixels can be obtained through resampling from the surface fitting the 3D projection points of the available pixels. To keep the latent geometrical structure between adjacent pixels, we modify the graph-regularized which allows the edge weights negative to decompose the approximation matrix. Experimental results of image inpainting demonstrate the effectiveness of the proposed method compared with the representative methods in quantities.
KeywordsGraph-regularized NMF Image inpainting Hole filling Approximation matrix
This work was supported by the National Natural Science Foundation of China (61702310 and 61772322).
- 4.Li, S.Z., Hou, X., Zhang, H., Cheng, Q.: Learning spatially localized parts-based representation. Comput. Vis. Pattern Recogn. 1, 1–207 (2001)Google Scholar
- 5.Shen, B., Si, L., Ji, R., Liu, B.: Robust nonnegative matrix factorization via l1 norm regularization. In: IEEE International Conference on Image Processing, pp. 1204–2311 (2012)Google Scholar
- 6.Yuan, Z., Oja, E.: Projective nonnegative matrix factorization for image compression and feature extraction. In: Proceedings of 14th Scandinavian Conference on Image Analysis (SCIA), Springer, pp. 333–342 (2005)Google Scholar
- 9.Luo, P., Peng, J.Y., Guan, Z.Y., Fan, J.P.: Dual regularized multi-view non-negative matrix factorization for clustering. Neurocomputing 294(24), 1–11 (2017)Google Scholar