Constant dimensionality reduction for large databases using localized PCA with an application to face recognition

This paper aims to reduce the complexities such as computation and storage of the facial data much further as compared to the methods described by PCA and LDA whilst keeping the discriminatory information, which is achieved by using a modified PCA technique along with an idea involving `separation of classes´ similar to LDA. Furthermore the problem that, `reduced dimensionality´ ironically increases with a growing database, is solved. Additionally, the possibility of updating the facial database dynamically for facilitating the most recent capture of a person is concluded to be much more feasible.