In this paper, we propose a novel method named the Multiple Constrained Mutual Subspace Method which increases the accuracy of face recognition by introducing a framework provided by ensemble learning. In our method we represent the set of patterns as a low-dimensional subspace, and calculate the similarity between an input subspace and a reference subspace, representing learnt identity. To extract effective features for identification both subspaces are projected onto multiple constraint subspaces. For generating constraint subspaces we apply ensemble learning algorithms, i.e. Bagging and Boosting. Through experimental results we show the effectiveness of our method.
[Publication (Japanese) ]