The Gradient-Descent Maximum Entropy (GDME) is an unsupervised endmember extraction algorithm. It is designed with the assumption that the original image would contain some pure pixels in, for example, background area.
run "gdme_test.m"
Lidan Miao, Hairong Qi (hqi@utk.edu), EECS, University of Tennessee, Knoxville
L. Miao, H. Qi, Harold Szu, "A maximum entropy approach to unsupervised mixed pixel decomposition," IEEE Transactions on Image Processing, 16(4):1008-1021, April 2007.
The conference version of this paper received the Best Paper Award at the International Conference on Patter Recognition (ICPR), 2006.