Unsupervised feature selection based on the two-dimensional principal component analysis and bipartite graph for face image classification

In this paper, we propose a new matrix-based feature selection method, called UFS۲DPCA, which leverages the hidden knowledge in orthogonal features obtained from two-dimensional principal component analysis (۲DPCA) to perform accurate unsupervised feature selection. The UFS۲DPCA algorithm first uses ۲DPCA to directly extract uncorrelated and orthogonal features from the two-dimensional image datasets. We then compute the correlation similarity between the main and extracted features. Finally, a weighted bipartite graph is constructed using two sets of features, and the best features are selected using the fast LAPJV algorithm. The selected features are classified using the K-Nearest Neighbor (KNN) classifier. To ensure statistical significance, the Friedman test is applied to compare the performance of UFS۲DPCA with other methods. The algorithm is evaluated on four well-known image datasets: Jaffe, Yale, ORL, and pixraw۱۰P. Key performance metrics such as accuracy, normalized mutual information (NMI), precision, recall, and F-measure are used for evaluation. The experimental results show that UFS۲DPCA consistently outperforms other state-of-the-art unsupervised feature selection methods. For example, UFS۲DPCA achieves an average NMI of ۰.۹۲۴۴ and average accuracy of ۰.۹۰۳۳ on the pixraw۱۰P face image dataset that has ۱۰۰۰۰ features. Similarly, it demonstrates superior performance in accuracy, recall, Precision, F-measure, and NMI across all datasets.

کلیدواژه ها:

Two-Dimensional Principal Component Analysis ، Weighted Bipartite Graph Matching ، LAPJV algorithm ، Augmenting Path ، similarity