Using Random Matrix Theory to determine the number of endmembers in a hyperspectral image

Abstract

Determining the number of spectral endmembers in a hyper-spectral image is an important step in the spectral unmixing process, and under- or overestimation of this number may lead to incorrect unmixing for unsupervised methods. In this paper we discuss a new method for determining the number of endmembers, using recent advances in Random Matrix Theory. This method is entirely unsupervised and is computationally cheaper than other existing methods. We apply our method to synthetic images, including a standard test image developed by Chein-I Chang, with good results for Gaussian independent noise.