An Experimental Study on the Effects of Noise on Endmember Extraction Methods

Abstract

Endmember extraction is frequently adopted to detect spectrally unique signatures of pure materials in the scene of hyperspectral imagery (HSI). In this paper, we investigate the effects of noise on seven widely used endmember extraction methods. We consider both Gaussian white noise and shot noise in our experiments. Experiments demonstrate that all methods considered in this paper are relatively robust to noise. Furthermore, all endmember spectral curves are contaminated by noise when noise is present in the hyperspectral data cubes. As a result, noise has a very important effect on the quality of the extracted endmember spectral curves.