A new endmember extraction algorithm by maximum ASC fraction

This paper presents a new simplex-based method for unsupervised endmember extraction, called maximum abundance sum-to-one constraint (ASC) fraction (MAF). The ASC fractions refer to the spectral unmixing results with the abundance sum-to-one constraint unmixing only. The algorithm assumes the existence of the pure pixels in the input data for every endmember in the scene, and exploits the fact that pixels with maximum ASC fractions are corresponding to the vertices of a simplex. In order to demonstrate the performance of the proposed MAF, the N-findr algorithm (N-FINDR) and vertex component analysis (VCA) based merely on PCA dimensional reduction are used for comparison. Experiments using both simulated and real hyperspectral data show that MAF is effective in searching optimal results, with a low computational complexity.