Orthogonal Graph Regularized Nonnegative Matrix Factorization for Image Clustering
Since high-dimensional data can be represented as vectors or matrices, matrix factorization is a common useful data modeling technique for high-dimensional feature representation, which has been widely applied in feature extraction, image processing and text clustering. Graph regularized nonnegative matrix factorization (GNMF) incorporates the non-negativity constraint and manifold regularization simultaneously to achieve a parts-based meaningful high-dimensional data representation, which can discover the underlying local geometrical structure of the original data space. In order to reduce the redundancy between bases and representations, and enhance the clustering power of NMF, three orthogonal variants of GNMF are proposed, which incorporates the orthogonal constraints into GNMF model. The optimization algorithms are developed to solve the objective functions of Orthogonal GNMF (OGNMF). The extensive experimental results on four real-world face image data sets have confirmed the effectiveness of the proposed OGNMF methods.
KeywordsImage representation Nonnegative Matrix Factorization Manifold regularization Orthogonal projection
This work was partially supported by National Natural Science Foundation of China (61902339, 61602388), China Postdoctoral Science Foundation (2018M633585, 2017M613216), Natural Science Basic Research Plan in Shaanxi Province of China (2018JQ6060, 2017JM6059), Fundamental Research Funds for the Central Universities (2452019064), Key Research and Development Program of Shaanxi (2019ZDLNY07-06-01), and the Doctoral Starting up Foundation of Yan’an University (YDBK2019-06).
- 3.Lam, E.Y.: Non-negative matrix factorization for images with Laplacian noise. In: IEEE Asia Pacific Conference on Circuits and Systems, pp. 798–801 (2008)Google Scholar
- 5.Kong, D., Ding, C., Huang, H.: Robust non-negative matrix factorization using L21-norm. In: Proceedings of the 20th ACM International Conference on Information and Knowledge Management, pp. 673–682 (2011)Google Scholar
- 6.Gao, H., Nie, F., Cai, W., Huang, H.: Robust capped norm nonnegative matrix factorization. In: Proceedings of the 24th ACM International Conference on Information and Knowledge Management, Melbourne, Australia, 19–23 October 2015Google Scholar
- 16.Ding, C., Li, T., Peng, W., Park, H.: Orthogonal nonnegative matrix t-factorizations for clustering. In: Proceedings of the 12th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 126–135. ACM (2006)Google Scholar