[Insight-developers] RE: ConfidenceConnectedImageFilter, MeanImageFunction, VarianceIm ageFunction

[Miller, James V (Research)]

The statistics inside are a little sloppy.  We treat the 
statistics from the seed region as population statistics
and define confidence regions.  The confidence regions
should broader to account for the fact that we are not using 
population statistics but rather estimates.

This is kind of a "poor man's version" of an algorithm in 
my dissertation.  The statistics there are more rigoruous
and take into account that the confidence intervals are
based on estimates.  Actually, I used prediction intervals
there that take into account how far the candidate points
are from the "seed" region in the domain (in other words,
how far the seed region is being extrapolated).





> -----Original Message-----
> From: Luis Ibanez [mailto:luis.ibanez@kitware.com]
> Sent: Thursday, November 14, 2002 11:38 AM
> To: Miller, James V (Research)
> Cc: 'Insight Developers'
> Subject: Re: ConfidenceConnectedImageFilter, MeanImageFunction,
> VarianceImageFunction
> 
> 
> Great,
> 
> I'll do the changes.
> the ConfidenceConnected filter has proved to
> be pretty useful.
> 
> Maybe you should came up with some complex
> statistical theory behind it and get it
> published. Users are asking for references
> to cite   :-)
> 
> 
>     Luis
> 
> 
> ==================================
> 
> 
> Miller, James V (Research) wrote:
> 
> > No issues here.  Are you going to make the changes?
> > 
> > I'd have the default radii be 1. But otherwise they sound 
> like good changes.
> > 
> > 
> > 
> 
>