文档介绍：$5REXVW)X]]\&OXVWHULQJ$OJRULWKPIRUWKH&ODVVLILFDWLRQRI5HPRWH6HQVLQJ,PDJHV
0DXUR%DUQL$QGUHD*DU]HOOL$OHVVDQGUR0HFRFFL/RUHQ]R6DEDWLQL
'HSDUWPHQWRI,QIRUPDWLRQ(QJLQHHULQJ8QLYHUVLW\RI6LHQD
9LD5RPD ±6LHQD,7$/<
7HO )D[
HPDLO ******@
ABSTRACT A new fuzzy clustering algorithm is proposed here, which,
by imposing a different constraint on the sum of memberships
A new fuzzy clustering algorithm is presented, that permits to across clusters, permits to assign a low membership to pixels
group data samples even when the number of clusters is not which do not belong to any clusters. Two different
known or when noise is present. The new algorithm is constraints are used alternately: the first corresponds to the
obtained by replacing the probabilistic constraint that probabilistic constraint and it applies to pixels whose
memberships across clusters must sum to one with a typicality as cluster members is high, whereas the second is
composite constraint. posite constraint allows the used for noisy pixels and outliers. In this way, pixels
algorithm to assign low memberships to uncertain data, thus extraneous to all clusters are easily recognized and not
ensuring higher robustness against noise, and avoiding the allowed to bias the final result. Experimental tests on a
need to know the nu