Spectral LWIR imaging for remote face detection

We show the utility of hyperspectral (HS) Longwave IR (LWIR) imaging for remote sensing human face detection. A proof of principle experimentation considers a limited but challenging dataset of calibrated LWIR HS data cubes for four skin tone diverse human subjects, standing outdoors at three distinct ranges. An algorithm is developed to capitalize on two spectral features suitable for small sample size targets, using all of the available bands. The algorithm maps the two spectral features-each consisting of sample sizes significantly smaller than the number of frequency bands, as it simultaneously generates two large sets (reference and testing) of independent contrasts in a lower dimensional subspace. The large sample size in the new subspace allows for the development of a strong hypothesis test that functions as a canonical target detector. Results using real HS imagery are encouraging for a specific face detection scenario, where the range is assumed to be known a priori-200, 300, 400ft.