Applications of morphological processing to endmember extraction

Mathematical morphology is a non-linear technique for spatial image analysis that has found many applications in different areas. This chapter reports on the extension of morphological image processing to hyperspectral imagery. In order to define extended morphological operations, a physically meaningful distance-based vector organization scheme is introduced, and fundamental vector operations are defined by extension. A specific application of extended morphological transformations is explored in this chapter: the Correspondence/Reprint request: Dr. Antonio J. Plaza, Computer Science Department, University of Extremadura, Avda. de la Universidad s/n, 10071 Caceres, Spain. E-mail: [email protected] Antonio J. Plaza 196 morphological identification of pure spectral constituents (endmembers) for mixed pixel characterization. Multi-channel morphological transformations demonstrate excellent performance when compared to other hyperspectral analysis methodologies in this particular application. A quantitative and comparative performance study in relation to available techniques, using real hyperspectral imagery collected by the 224-channel NASA/Jet Propulsion Laboratory Airborne Visible-Infrared Imaging Spectrometer (AVIRIS), reveals that the complementary utilization of spatial and spectral information by the proposed transformations alleviates the problems related to each of them taken separately. The chapter also develops a parallel implementation that speeds up computational performance.