A Mathematical Morphology Based Approach to Locating Spectral Endmembers

Spectral endmembers are needed in order to apply Spectral Mixture Analysis (SMA) based on imaging spectrometer data. Currently there are several techniques able to find image endmembers in N spectral dimensions and are wholly based on the spectral feature space. However, modelling only the outside hull of the spectral scatter provides only a basic understanding of the spectral variability. Our ultimate goal is to model the n-dimensional spectral space from the inside, out thereby providing a map of image spectral objects. The research presented in this paper are initial findings applying mathematical morphology to imaging spectrometer data. Mathematical morphology was applied in two ways: i) by measuring the morphological hyperspectral scalar gradient at each pixel position using a structuring element of predefined size and shape and ii) by determining whether this pixel falls within the smallest enclosing hyperbox containing the pixels belonging to the structuring element centred at this position. While the results are interesting in themselves, they were unable to automatically provide the image endmembers. However, the results provide a wealth of information on the local variations in spectral profiles which will be used to model their behaviour within the spectral feature cloud.

BIELSKI Conrad;  SOILLE Pierre; 

2006-01-20

EARSEL & WARSAW UNIVERSITY

JRC32040