A novel information-theoretic measure for face frequency feature selection

Feature selection aims to reduce the dimensionality of patterns for classificatory analysis by selecting the most informative instead of irrelevant and/or redundant ones. In the case of a face recognition problem, the objective of feature selection is to find the smallest subset of features that maximizes the face recognition ability. The drawback of the most recently proposed feature selection based information measures is that they do not take into consideration the interaction between features. Indeed, a single feature can be considered irrelevant based on its correlation with the class; but when combined with other features, it becomes very relevant. Unintentional removal of these features can result in a loss of useful information and thus may cause poor recognition performance. In this paper, we propose a novel information-theoretic measure for face frequency feature selection, named Ponderated Mutual Information (PMI), that takes into account the interaction between features. Significant performances are obtained when applying this method on a benchmark face databases.