Melanosome level estimation in human skin from hyperspectral imagery

Abstract

Locating individuals in the open has several practical uses; most formidable is that of the search and rescue application. Although existing methods exist to find human skin in color imagery, these methods are subject to high false alarm rates caused by objects that are skin colored. Hyperspectral imagery offers a distinct advantage due to the abundance of spectral information that can be exploited to dramatically reduce false alarms while maintaining a high detection rate. The work presented in this article extends our earlier work in hyperspectral-based skin detection to the detection of skin pigmentation levels. Specifically, we estimate the amount of melanosomes contained within pixels identified as skin which gives an estimate of skin color. Our method is based on the intrinsic properties of human skin and does not use a “hyperspectral to RGB conversion.” We demonstrate the capability of our algorithm using a hyperspectral instrument developed by SpecTIR Corp (the HST3) which nominally covers the spectral range of 400–2500nm.