Using principal component analysis for endmember extraction

This paper introduces a new simplex-based unsupervised endmember extraction method from hyperspectral data. The method exploits the dimensionality reduction ability of the principal component analysis, and generalizes the concept that, the first generated endmember by the Simplex Growing Algorithm, is always a pixel which has either a maximum or a minimum value in the first component, to more endmembers and components. According to the method, a subset of the minimum and maximum values of the first p-1 principal components, where p is the number of the endmembers to be defined, corresponds to the vertices of the simplex which is created by the data. In order to evaluate the proposed method, simulated images with different noise levels were created. For comparison purposes, several other known endmember extraction algorithms were applied to the data and compared with the new method. Results present that the proposed method can be promising in the field of endmember extraction.