| Re: Fuzzy classification [message #39449 is a reply to message #39336] |
Tue, 18 May 2004 05:08  |
Vincenzo Positano
Messages: 2
Registered: January 2001
|
Junior Member |
|
|
"Joao F." <joao.antunes@agr.unicamp.br> ha scritto nel messaggio
news:7568d93c.0405140734.5b3e4df4@posting.google.com...
> Hello,
>
> I am looking for an IDL implementation of the fuzzy c-means (FCM)
> clustering algorithm and fuzzy maximum likelihood algorithm (FML) to
> classication of satellite images. The resulted cluster centers
> obtained by the FCM algorithm from the first step are input into the
> FML algorithm.
>
This is an implementation of FCM for image segmentation.
mu initialization should be adapted to your problem.
The function seems to work on medical images with 3-4
classes.
Vincenzo Positano
;image segmentation based on fuzzy clustering
;
; image input image
; num_class number of classes
; m fuzzy factor
; eps threeshold
; n_max max iteration number
; a cluster centers
function fuzzy_clustering,image,num_class,m,eps,n_max,CENTER=a
sz=SIZE(image,/DIMENSIONS)
x = REFORM(image,sz[0]*sz[1])
n=LONG(N_ELEMENTS(x))
c=num_class
a=DBLARR(c)
mu=DBLARR(c,n)
kk=0
;mu initialization
idx=WHERE(image le 20)
mu[0,idx]=1.0
idx=WHERE((image ge 20) and (image lt 200))
mu[1,idx]=1.0
idx=WHERE(image ge 200)
mu[2,idx]=1.0
mu_old=DBLARR(c,n)
me=(2/(m-1.0))
while (kk le n_max) do begin
for i=0L,c-1 do begin
tmp=REFORM(mu[i,*],n)
num = tmp^m*x
den = tmp^m
a[i]=TOTAL(num)/TOTAL(den)
endfor
for i=0L,c-1 do begin
num = DBLARR(c,n)
den = DBLARR(c,n)
num[0,*]=(ABS(x-a[i]))^me
for k=0L,c-1 do begin
den[k,*]=(ABS(x-a[k]))^me
num[k,*]=num[0,*]
endfor
pippo=TOTAL(num/den,1)
mu[i,*]=1.0/pippo
endfor
diff = TOTAL(ABS(mu-mu_old))/FLOAT(n)
if (diff le eps) THEN KK=n_max
mu_old=mu
kk=kk+1
endwhile
return,mu
end
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
