Palmprint recognition with an efficient data driven ensemble classifier

Palmprint recognition is an important and widely used modality in biometric systems. It has a high reliability, stability and user acceptability. This paper proposes a new and effective ensemble classifier for palmprint recognition based on Random Subspace Method (RSM). The method relies on 2DPCA to build nearly incoherent random subspaces. As 2DPCA is an unsurpevised technique, features are extracted in each subspace using 2DLDA. A simple 1-Nearest Neighbor classifier is associated to each subspace, the final decision rule being obtained by a majority voting rule. Extensive experiments on three public palmprint datasets have been conducted to compare the proposed approach to existing methods. The experimental results demonstrate that our method improves on the state-of-the-art. It turns out that for this kind of data, the use of weak classifiers learned over nearly incoherent features is very efficient. Besides these findings, we provide an empirical analysis of the parameters involved in the random subspace technique to guide the user in the choice of the appropriate hyper-parameters.