Locality sensitive semi-supervised feature selection

In many computer vision tasks like face recognition and image retrieval, one is often confronted with high-dimensional data. Procedures that are analytically or computationally manageable in low-dimensional spaces can become completely impractical in a space of several hundreds or thousands dimensions. Thus, various techniques have been developed for reducing the dimensionality of the feature space in the hope of obtaining a more manageable problem. The most popular feature selection and extraction techniques include Fisher score, Principal Component Analysis (PCA), and Laplacian score. Among them, PCA and Laplacian score are unsupervised methods, while Fisher score is supervised method. None of them can take advantage of both labeled and unlabeled data points. In this paper, we introduce a novel semi-supervised feature selection algorithm, which makes use of both labeled and unlabeled data points. Specifically, the labeled points are used to maximize the margin between data points from different classes, while the unlabeled points are used to discover the geometrical structure of the data space. We compare our proposed algorithm with Fisher score and Laplacian score on face recognition. Experimental results demonstrate the efficiency and effectiveness of our algorithm.