Graph Manifold Clustering based Band Selection for Hyperspectral Face Recognition

Efficient band selection reduces the computational load associated with the processing of hyperspectral data. In this paper we propose a novel graph manifold approach based band selection framework for hyperspectral face recognition. In this work, we extract facial features by local binary pattern, which is a popular facial feature descriptor. In the next stage, we cluster the data points using graph manifold ranking method and select representative bands from each cluster. We also propose a band similarity index (BSI) for quantifying the consistency of band selection algorithms. BSI facilitates faster and efficient matching of HSI faces provided the selected bands capture inter and intra-subject variation. We compare the efficiency of the proposed framework with state of the art methods on two real hyperspectral face datasets.