Improved Multispectral Skin Detection and its Application to Search Space Reduction for Dismount Detection Based on Histograms of Oriented Gradients [open pdf - 5MB]
From the thesis abstract: "Due to the general shift from conventional warfare to terrorism and urban warfare by enemies of the United States in the late 20th Century, locating and tracking individuals of interest have become critically important. Dismount detection and tracking are vital to provide security and intelligence in both combat and homeland defense scenarios including base defense, combat search and rescue (CSAR), and border patrol. This thesis focuses on exploiting recent advances in skin detection research to reliably detect dismounts in a scene. To this end, a signal-plus-noise model is developed to map modeled skin spectra to the imaging response of an arbitrary sensor, enabling an in-depth exploration of multispectral features as they are encountered in the real world for improved skin detection. Knowledge of skin locations within an image is exploited to cue a robust dismount detection algorithm, significantly improving dismount detection performance and efficiency. This research explores multiple spectral features and detection algorithms to find the best features and algorithms for detecting skin in multispectral visible and short wave infrared (SWIR) imagery. This study concludes that using SWIR imagery for skin detection and color information for false alarm suppression results in 95% probability of skin detection at a false alarm rate of only 0.4%."
Air Force Institute of Technology Report No. AFIT/GE/ENG/10-05
Defense Technical Information Center (DTIC): http://www.dtic.mil/dtic/