Endmember extraction by exemplar finder

We propose a novel method called exemplar finder (EF) for spectraldata endmember extraction problem, which is also known as blind unmixing inremote sensing community. Exemplar finder is based on data self reconstructionassuming that the bases (endmembers) generating the data exist in the givendata set. The bases selection is fulfilled by minimising a l2=l1 norm on the reconstructioncoefficients, which eliminates or suppresses irrelevant weights fromnon-exemplar samples. As a result, it is able to identify endmembers automatically.This algorithm can be further extended, for example, using different errorstructures and including rank operator. We test this method on semi-simulatedhyperspectral data where ground truth is available. Exemplar finder successfullyidentifies endmembers, which is far better than some existing methods, especiallywhen signal to noise ratio is high.