AIMS shape moment on disconnected objects in same volume

AIMS library and commands usage

Moderators: denghien, riviere

Post Reply
Posts: 1
Joined: Mon Jun 04, 2007 9:43 am

AIMS shape moment on disconnected objects in same volume

Post by shishir »

Hi Dr. Mangin,

I wanted to ask you a fairly basic question about your BrainVisa program, more specifically with the AIMS shape moment command. If you have a binary volume of 3 disconnected objects (with the same label), how does aims/vip calculate the moment features? Does it calculate each object separately and then find the shape moments/moment invariants and subsequently averages it to output one table of features? Is there a way to calculate the shape properties separately without changing the label for each of the 3 disconnected objects?

Also, in your brain morphometry analysis with 3D shape moments you only took into consideration the invariant moment features...not the other shape moment features (i.e. nu2xy, etc) right? And the reasoning was that the other shape moment features would have a large degree of variation for a same object that was had been modified via translation/rotation?

Thanks again,

User avatar
Site Admin
Posts: 1361
Joined: Tue Jan 06, 2004 12:21 pm
Location: CEA NeuroSpin, Saint Aubin, France

Post by riviere »

Which program are you talking about ? the AimsMoment command, or is there anything in brainvisa ?
For disconnected objects both behaviours are possible, but we normally calculate moments on the union of all parts to have a homogen set of features even if objects you are comparing have different numbers of connected components.
We look at invariant moments to try to study 'pure shape' descriptors and see if there is anything significant in it. Position, size and orientation are extracted separately, and so the moment features are not correlated to them. Moreover position, size and orientation might be correlated to global effects (size of the subject, or even registration errors) so the invariant moments are 'cleaned' of these potential correlations.
User avatar
Jean-Francois Mangin
Posts: 337
Joined: Mon Mar 01, 2004 10:24 am
Location: Neurospin, CEA, France

Post by Jean-Francois Mangin »

If you labelled the different objects differently, I think you can select the object you want to compute the invariant for. I am traveling this week, I can not tell you more. Furthermore, the guy who coded all this stuff is Fabrice Poupon, hope he will tell you more...

You are right, we do not use the nu stuff
Post Reply