Abstract:
Heterogeneous biometric recognition requires matching images with variations such as resolution and spectrum. Heterogeneity in images often reduces the inter-class homogeneous distance while increasing the intra-class heterogeneous distance. In this research, a novel metric learning method is proposed, which minimizes the intra-class homogeneous and heterogeneous distances while maximizing the inter-class homogeneous and heterogeneous distances. The effectiveness of the proposed algorithm is demonstrated on three face databases and three periocular databases
corresponding to real-world heterogeneous biometric recognition problems. The experiments show that the proposed algorithm provides state-of-the-art results on all the databases and outperforms existing recognition and metric learning algorithms. Further another use case is presented for mobile periocular recognition, proving the efficacy of the proposed model for unconstrained settings.
