H-COMP: a tool for quantitative and comparative analysis of endmember identification algorithms

Abstract

Over the past years, several endmember extraction algorithms have been developed for spectral mixture analysis of hyperspectral data. Due to a lack of quantitative approaches to substantiate new algorithms, available methods have not been rigorously compared using a unified scheme. In this paper, we describe H-COMP, an IDL (Interactive Data Language)-based software toolkit for visualization and interactive analysis of results provided by endmember selection methods. The suitability of using H-COMP for assessment and comparison of endmember extraction algorithms is demonstrated in this work by a comparative analysis of three standard algorithms: Pixel Purity Index (PPI), N-FINDR, and Automated Morphological Endmember Extraction (AMEE). Simulated and real hyperspectral datasets, collected by the NASA/JPL Airborne Visible-Infrared Imaging Spectrometer (AVIRIS), are used to carry out a comparative effort, focused on the definition of reliable endmember quality metrics.