Image-level Classification in Hyperspectral Images using Feature Descriptors, with Application to Face Recognition

Image-level classification from hyperspectral images (HSI) has seldom been addressed in the literature. Instead, traditional classification methods have focused on individual pixels. Pixel-level HSI classification comes at a high computational burden though. In this paper, we present a novel pipeline for classification at image-level, where each band in the HSI is considered as a separate image. In contrast to operating at the pixel level, this approach allows us to exploit higher-level information like shapes. We use traditional feature descriptors, i.e. histograms of oriented gradients, local binary patterns, and the scale-invariant feature transform. For demonstration we choose a face recognition task. The system is tested on two hyperspectral face datasets, and our experiments show that the proposed method outperforms the existing state-of-the-art hyperspectral face recognition methods.