Handling spectral variability with alternating angle minimization

We present an algorithm for solving the spectral mixture analysis problem in situations that show spectral variability, represented by spectral libraries for each class of interest. The aim is to select a single optimal endmember candidate from each library, on a per-pixel basis. The algorithm exploits the high dimensionality of spectral data, and is based on an alternating spectral angle minimization scheme. The resulting algorithm is computationally very efficient, and yields excellent results compared to other techniques, such as multiple endmember spectral mixture analysis.