Automated labeling of segmented hyperspectral imagery via spectral matching

Despite recent advances in hyperspectral image processing, automated material identification from hyperspectral image data is still an unsolved problem. In this work, we develop a technique for labeling hyperspectral imagery, which leverages segmented image data and a library of spectral signatures of materials. We define a new spectral similarity measure that considers continuum removed spectra in addition to continuum intact reflectance spectra. We show that using both of these characteristics in similarity analysis yields improved results over recently proposed similarity measures. Analysis on an AVIRIS image of an urban scene is presented.