[Insight-users] Registering Low Contrast Images

[Vikren Sarkar]

Greetings, 
 
I had posted a question a while back about intensity-based registration for low-contrast images. Thank you Luis for providing some pointers. You had requested I post example images so that you and the rest of the community might suggest what metrics are best suited to try to do automatic registration. I was waiting until I was able to create my own images before I posted them. The following links show two images, one which is rotated 5 degrees with respect to the other (rotation about the image center). 
 



http://www.hotlinkfiles.com/view/full/1600023_zwyje http://www.hotlinkfiles.com/view/full/1600024_ujxkr
 
These images are part of 3D-tomographic datasets. I chose the two corresponding slices that showed the most contrast to post as an example. All other images have even less contrast than shown here. 
 
Bearing in mind that I am new to ITK, I was wondering whether someone could suggest a metric that would be good for automatically registering two 3D datasets of such images. I would be happy to post the full image-sets if that would be more helpful. I am trying to determine the shifts in the three axes as well as rotations about each axis. 
 
Any help that anybody can provide will be much appreciated. 
 
Vik> > Hi Vikren,> > > 1) Yes, it is possible with ITK to compute the rigid transform that> will map the coordinate system of your fixed image into the> coordinate system of the moving image.> > > 2) Could you post screeshots of your low contract images in a public> web site ?> > Being able to see the images will help us a lot for recommending> you an image metric to use.> > > In the absence of being able to see your image, here are two> suggestions that come to mind:> > A) Normalized Correlation metric> > B) Passing your images through edge detection filters> and then using the results as the actual images to> register. (you will find several edge-detection> filters in ITK, you probably want to start with> a simple gradient magnitude filter).> > > C) Mutual information is still an option that you> could reconsider, and for which you may have to> adjust the parameters of number of bins and> number of samples.> > > > In general, you should be careful discarding a particular technique> based on having tried it once, with only a reduced set of parameters.> > It is not uncommon that for a fine-tunning exercise to be required> before you get acceptable results with a given technique.> > > > Regards,> > > Luis> > > ---------------------> Vikren Sarkar wrote:> > Greetings,> > > > > > > > I am new to ITK and am trying to use the registration algorithms to> > register two 3D x-ray image datasets. The problem is that all the> > images have very low contrast and the intensity-based algorithms> > fail when used with the mean square metric (I also tried mutual> > information with no success). I need to perform a rigid registration> > and am hoping someone with more experience could suggest a metric> > which would be appropriate for use with low contrast images.> > > > > > > > Alternatively, I am open to the idea of using control points which> > the user would input. I am most interested in the actual values of> > translation, rotation and scaling required for the registration than> > in the final registered datasets themselves. The reason is that we> > are investigating the use of those images in patient positioning> > prior to radiation delivery. Can someone please point me in the> > right direction as to whether this is possible using ITK? It is> > probably obvious that I am very new at this and I would appreciate> > any help and/or pointers. Thank you.> > > > > > > > Vik> > 
_________________________________________________________________
Time for vacation? WIN what you need- enter now!
http://www.gowindowslive.com/summergiveaway/?ocid=tag_jlyhm
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20080717/18a07d64/attachment.htm>