Visible-light and near-infrared face recognition at a distance

A method to solve the problem of face recognition at a distance (FRAD) under the visible-light (VIS) and the near-infrared (NIR) spectra is presented in this work. For images taken under visible light at day time, we perform the coarse-scale alignment/enhancement to eliminate a set of unlikely candidates at the first stage. Then, the fine-scale alignment/enhancement steps are conducted to refine the candidate list furthermore iteratively at the second stage. To address the additional challenge associated with NIR images captured at night time, we incorporate a restoration mechanism that reconstructs low-quality patches through a locally linear embedding (LLE) process with a local constraint. It is shown by experimental results that our FRAD solution outperforms state-of-the-art methods on both VIS and NIR images.