Fuzzy c-means thresholding

FCMTHRESH Thresholding by 3-class fuzzy c-means clustering
[bw,level]=fcmthresh(IM,sw) outputs the binary image bw and threshold level of image IM using a 3-class fuzzy c-means clustering. It often works better than Otsu's methold which outputs larger or smaller threshold on fluorescence images.

sw is 0 or 1, a switch of cut-off position.
sw=0, cut between the small and middle class
sw=1, cut between the middle and large class

Contributed by Guanglei Xiong (xgl99@mails.tsinghua.edu.cn)
at Tsinghua University, Beijing, China.

Inspired by Xiaobo Zhou
at Harvard Medical School.