In this paper, we proposed a new linear regression-based approach for face recognition, called farthest subspace classification. In previous literatures, it was believed that the facial images from a specific object class tend to lie on a linear subspace, i.e. the class-specific subspace. Therefore a query image will be considered belonging to its nearest subspace (NS) of a class. The distance from a query image to each class-specific subspace is calculated simply by the linear regression. In this paper, we proposed a novel notion of face recognition that in the complete feature space spanned by all the gallery images, each class-specific subspace has not only common subspace shared by every class-specific subspace, but also its unique coordinate bases, which are available discriminative information. Based on this notion, we develop farthest subspace (FS) classifier to perform face recognition. The experimental results supported the proposed novel concept. Furthermore, we proposed nearest-farthest subspace (NFS) classification using both NS and FS rules, which outperform NS used alone. © 2012 Springer-Verlag.
CITATION STYLE
Mi, J. X. (2011). A new subspace approach for face recognition. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 6840 LNBI, pp. 551–557). https://doi.org/10.1007/978-3-642-24553-4_73
Mendeley helps you to discover research relevant for your work.