[Insight-users] itkExpectationMaximizationMixtureModel

[Jisung Kim]

Hi ahmad,

--- ahmad bukhari <highway_111@mail.com> wrote:
> hi  jisung and others... 
> The assignment of weightages is global and hence it
> shall not be effective 
> ..In a certain intensity zone a class may not be
> present at all and then 
> assigning weightages does not help in any way.... 

No, it wouldn't be helpful. However, if a component is
correctly implemented, then it wouldn't hurt the
result neither, because the components' Evaluate
method should return very very small density value in
that intensity zone.

> An example of adding a component shall be useful for
> me. 
> thanks 
> ahmad 
> ----- Original Message -----  
> From: Jisung Kim <bahrahm@yahoo.com>  
> Date: Fri, 25 Oct 2002 07:11:31 -0700 (PDT)  
> To: ahmad bukhari <highway_111@mail.com>, insight  
> <insight-users@public.kitware.com>  
> Subject: Re: [Insight-users]
> itkExpectationMaximizationMixtureModel  
>   
>   
> > Hi ahmad,  
> >   
> >   
> > --- ahmad bukhari <highway_111@mail.com> wrote:  
> > >  hi everyone!!   
> > > There is another issue that i need help with:   
> > > the mixture model can finally give you an
> estimate  
> > > of   
> > > the different classes presence....but the
> trouble is  
> > >   
> > > that different classes occur in different
> intensity   
> > > regions. For example in an intensity zone on  a 
> 
> > > typical   
> > > frequency histogram plot different classes occur
> in   
> > > different regions interfering in some cases and 
> 
> > > isolated   
> > > in rest. So the mixture model perhaps should be 
> 
> > > applied   
> > > (initialized)differently for different intensity
>  
> > > zones   
> > > to get an accurate picture. Is there a way
> out...??   
> >   
> > Each iteration of the EM estimation process
> updates  
> > weights for measurement vectors with given  
> > "components". In other words, in the test program,
>  
> > each measurement vector has two weight values that
>  
> > tells you how likely the measurement vector
> belongs to  
> > the two Gaussian classes. When the Gaussian
> components  
> > update  their parameters, they use the weights
> instead  
> > of using simple mean and covariance calculations.
> This  
> > weighted mean and weighted covariance calculations
>  
> > actually has similar effect that you mentioned. 
> If a  
> > measurement vector highly likely belong to a
> component  
> > (let's say component A) but not to  the other  
> > component (component B), then the measurement
> vector  
> > will affects the parameter calculation of
> component A  
> > but won't affect much that of component B.  
> >   
> > Is this what you need?  
> >   
> >     
> >   
> > > thanks   
> > > ahmad     
> > > --   
> > >   
> >   
> > Thanks,  
> >   
> > Jisung.  
> >   
> >
>
__________________________________________________________
>  
> > > Sign-up for your own FREE Personalized E-mail at
>  
> > > Mail.com  
> > > http://www.mail.com/?sr=signup  
> > >   
> > > _______________________________________________ 
> 
> > > Insight-users mailing list  
> > > Insight-users@public.kitware.com  
> > >  
> >
>
http://public.kitware.com/mailman/listinfo/insight-users
>  
> >   
> >   
> > =====  
> > Jisung Kim  
> > bahrahm@yahoo.com  
> > 106 Mason Farm Rd.  
> > 129 Radiology Research Lab., CB# 7515  
> > Univ. of North Carolina at Chapel Hill  
> > Chapel Hill, NC 27599-7515  
> >   
> > __________________________________________________
>  
> > Do you Yahoo!?  
> > Y! Web Hosting - Let the expert host your web site
>  
> > http://webhosting.yahoo.com/  
> >   
> -- 
>
__________________________________________________________
> Sign-up for your own FREE Personalized E-mail at
> Mail.com
> http://www.mail.com/?sr=signup
> 


=====
Jisung Kim
bahrahm@yahoo.com
106 Mason Farm Rd.
129 Radiology Research Lab., CB# 7515
Univ. of North Carolina at Chapel Hill
Chapel Hill, NC 27599-7515

__________________________________________________
Do you Yahoo!?
Y! Web Hosting - Let the expert host your web site
http://webhosting.yahoo.com/