[Insight-users] Is this a registration problem?

[Martin Urschler]

hello sotiris,

Sotiris Dimopoulos wrote:
> I have created a 3d object from ct slices and i would
> like to perform a method that will "calculate" the
> parameters of a 3d geometrical model(such as a
> cylinder)that will estimate - in an optimal way- my
> initial reconstruced model.I suppose that the
> radius,height and orientation of a spatial object will
> be parameters. 
> 
> Is this a "model to image registration" problem? 
> Have you got any ideas to start working with?

IMHO what you are describing is a model fitting problem. There are 
different approaches that you could look into. Two simple ones that 
immediately come into my mind is to take a parametric model (e.g. your 
cylinder, or a general quadric surface) and use your model points in a 
least squares fitting routine, or you could take all your model points 
and perform a principal component analysis to get principal axes which 
would correspond to an ellipsoid model. There are more elaborate models 
like e.g. superquadrics. I guess by googling for "3d model fitting" or 
"3d surface fitting" you should get some nice links, also the CVonline 
web page may be a good starting point.

As for using the spatial objects of ITK for this purpose, I am not a 
competent person to answer if it is possible to do that, since I didn't 
look into this framework yet.

hth,
Martin