Hyperspectral face data reduction and classification with multiresolution wavelets

Abstract

Hyperspectral face images present productive information captured using a Hyperspectral camera compared to normal RGB camera capturing face images. Hyper spectral imaging is the collecting and processing of information from across the visible electromagnetic spectrum. Hyper spectral imaging deals with the imaging of narrow spectral bands over a continuous visible spectral range, and produces the spectra of all pixels in the scene. In this research face recognition experimentation is done in near infrared Hyperspectral images. The recognition is accomplished on Hyperspectral face database consisting of 47 test subjects created by Hong Kong Polytechnic University. The database images of Hyperspectral faces were collected using a CCD camera equipped with a liquid crystal tunable filter to provide 33 bands over the near-infrared (0.7_m-1.0_m). Experiments were conducted to demonstrate that this simple algorithm can be used to recognize faces that changes in facial pose and expression over time.