[Insight-users] Image Registration

[Vikash Gupta]

Yes, I checked now, indeed one of the images has a lot of empty background, so, it kind of makes sense, but I will still verify everything with respect to the center of the image as you mentioned. 
Finally now I can go ahead and write my codes, I was really stuck 
. 
Thanks 
Vikash 

----- Original Message -----

> From: "Torsten Rohlfing" <torsten at synapse.sri.com>
> To: "Vikash Gupta" <vikash.gupta at inria.fr>
> Cc: "Vikash Gupta" <write2vikash at gmail.com>, insight-users at itk.org
> Sent: Friday, September 14, 2012 12:33:46 AM
> Subject: Re: [Insight-users] Image Registration

> On 09/13/2012 03:27 PM, Vikash Gupta wrote:

> > Hi Torsten,
> 
> > Thanks once again for your patient email. Yes so I resampled the
> > moving image to the fixed image space, with the output
> > transformation matrix (my output transformation matrix is inverted)
> > and I get the image in the expected co-ordinates of the fixed
> > image.
> 
> > So, I will apply this transformation (or the inverse) to the points
> > in physical spaces of the images in order to take one to other or
> > the way around, as you suggested.
> 

> > One last thing that startles me as i mention before is the gap
> > between the translation parameters, which intutively should be
> > more,
> > but I didn't write the registration routine so probably that is an
> > issue I should find out by looking at the registration routine, how
> > things are implemented there.
> 
> Well, one explanation for the near-zero translation despite very
> different origins could be if the images had very different fields
> of view.

> If you simply add a lot of background around one of the images
> without changing the physical coordinates of the pixels that are
> already there, then you will need to adjust the image origin
> accordingly, but the translation vector would not change at all.

> The point here is, that the first pixel in your image (which is the
> origin pixel) rarely has any meaningful anatomy information, so
> where it maps matters little to nothing. It is usually more
> interesting to see where the CENTER pixel maps; I suspect in your
> images, the center of each is near the zero physical coordinate. So
> if you have your object of interest (e.g., patient) centered in your
> images, and the center is also the zero coordinate, then there would
> be little translation needed to align them. Yet, if the images have
> different fields of view, your origins would be very different.

> And you're very welcome :)

> TR
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20120914/1df303b9/attachment.htm>