Fuzzy K-Nearest Neighbor Graph Regularized Non-Negative Matrix Factorization

Non-negative matrix factorization (NMF) is a feature extraction algorithm for extracting features from non-negative data, but NMF puts more emphasis on local information and ignores global information, which leads to limitations in classifying noisy images. To solve this problem, a new method called Fuzzy Graph Regularized Non-Negative Matrix Factorization (FKGNMF) is proposed to integrate fuzzy set theory into the NMF framework. By combining with graph embedding theory, this method can take into account the properties of the global structure of the image, mark the membership degree of the sample with the fuzzy membership degree, and provide fuzzy decisions based on fuzzy labels. The locality and globality of the data are preserved, improving the robustness and accuracy of feature extraction. Experimental evaluations on various bench-mark datasets show that FKGNMF outperforms traditional NMF and other mainstream methods. Compared to other algorithms, this method can increase the recognition rate by 2%-10%.