Registering distorted images

Frank Valdes wrote on Jul 27, 1999
>From ericr@astro.yale.edu Fri Jul 23 09:54:23 1999
Date: Fri, 23 Jul 1999 11:49:32 -0400 (EDT)
From: "Eric P. Rubenstein" <ericr@astro.yale.edu>
To: fvaldes@noao.edu, kaiser@ifa.hawaii.edu,
        Eric Rubenstein <ericr@astro.yale.edu>,
        "Gil'ad N. Cohen" <Cohen_Gilad/mskcc_MedPhys@mskmail.mskcc.org>
Subject: Registering distorted images

Dear Drs. Valdes and Kaiser,

I am a post-doc at Yale and CTIO (located at the latter this year). Nick
Suntzeff suggested that I write to both of you concerning both an
astronomy problem I am having and also regarding a medical imaging problem
a friend is having.  I think that my problem is easier, so I'll begin with
that one.

I am currently fighting my way through trying to cross-identify two sets
of stars taken from WFPC2 at different roll-angles.  I only have the
photometry tables from the older epoch (not my data) but have both data
and photometry for the recent epoch (my data).  I tried to use Peter
Stetson's DAOMASTER matching routines, but there is too much distortion of
the images relative to each other, so only a sub-set of the stars match.
Do you know if there is a more robust way to do star list matching (other
than by eye)?  I need to match the first epoch PC and WF2-4 fields with my
own, but because of the different roll-angle, some of their PC data is in
my WF data and vice versa.

That is it for my question, the medical imaging question is next:


---------- Forwarded message ----------
Date: Fri, 23 Jul 1999 10:15:49 -0400
From: Cohen_Gilad/mskcc_MedPhys@mskmail.mskcc.org
To: ericr@astro.yale.edu

<snip>
     
     The case involves MRI images but any image warping/morphing/whatever 
     algorithm would be welcome at this point.
     
     Take two MRI studies done on the same patient -- imaging his pelvis.  
     
     One uses a body coil gives which is a general portrayal of geometry of 
     the pelvis with no distortions whatsoever.  however it does not 
     provide the diagnostic information we need.
     
     The second uses  rectal coil (-no remarks please!) which inflated with 
     100 cc of air.  In introducing it the prostate gland is deformed 
     (-obviously).  This study gives us particular regions of interest 
     within the prostate.
     
     When the patient finally gets treated the imaging information 
     resembles the body coil study.  how does one unwarp the prostate 
     information and register it from case 2 to case 1.
     
     Current state of affairs/Assumptions that can be made rather safely:
     
     1. When the prostate is being outlined  the outline may have some 
     error, but they are roughly ok -- this has to be done manually
     2. It is safe to assume that the total volume of the prostate remains 
     constant.
     3. we can co-register different imaging studies, but do not - as of 
     yet - have the capability of unwarping a section of the image.
     
     Any ideas?
     

--------------(end forwarded message)---------------

Any help on the astronomy question would be most appreciated by me--any
help on the second question might help save some lives.  Either way, thank
you for any help you can provide.

Eric



Eric P. Rubenstein          ericr@astro.yale.edu

CTIO
Casilla 603
La Serena, Chile

and

Astronomy Department
Yale Univ; Box 208101
New Haven, CT 06520-8101



>From valdes Tue Jul 27 11:19:58 1999
Date: Tue, 27 Jul 99 11:19:57 MST
From: valdes (Frank Valdes)
To: ericr@astro.yale.edu
Subject: Re: Registering distorted images

Hello Eric,

The approach I am familiar with for matching coordinate lists is based
on finding similar triangles.  The method is described in various places.
My implementation of this idea in FOCAS (a package for catalog generation
and analysis) is described at

F. G. Valdes, L. E. Campusano, J. D. Velasquez, and P.B. Stetson, 1995,
PASP, 107, 1119.

This method is, I believe, similar to what is in DAOPHOT from Peter Stetson
(though I don't know this for a fact).  IRAF has a algorithm like this.
However I don't know the name and the author of this is on vacation.
You could write Lindsey Davis (ldavis@noao.edu) asking about matching of
two lists of coordinates.  These methods assume the transformation is
linear with offsets, rotations, and possibly flips.  If there is some
additional significant distortion no general algorithm is likely to work.
Since this is WFPC data you can ask Help@stsci.edu about registering
WFPC data at different roll angles.

Other than that it may require interactively matching things.



Concerning your medical imaging question I can answer with IRAF tasks
that do a general unwarping.  The tasks are GEOMAP (to make the
transformation map) and GEOTRAN.  These task describes things in terms
of star positions but if you can measure fiducial points between the
two images that would be equivalent.  The transformations are expected
to be describable by 2D functions so there has to be some degree of
smoothness; i.e. no kinks and discontinuities.  However, the medical
case described might well be done this way.  If someone attempts this I
would also point out that measuring the fiducial points might best be
done using the task RIMCURSOR which simply reads the image display cursor
(the image data would have to be converted to a format that can be
displayed with IRAF display servers XIMTOOL or SAOIMAGE such as FITS
and displayed) and returns the pixel coordinates of marked points.

Cheers,
Frank Vlades


I am going to take the liberty of posting your message to the ADASS
(Astronomical Data Analysis Software and Systems) application newsgroup
where there is some readership that may be interested in the problems
and have some ideas.
Last post on Jul 27, 1999
IRAF Community Distribution

IRAF maintained by the community

Registering distorted images
