Thoughts on common image registration questions

[Pixel.to.life]

I saw quite a few questions on how to estimate a rigid/affine
transform between two given sets of images, assuming they display the
same rigid object in different poses.

Here is a compilation of some thoughts (purely for educational
purposes, do not take them as an oath:-). They are excerpts from what
I shared with other group-discussions before this groups was formed:

1. The most common term, I guess, would be 'image registration'. It
is
usually considered a multi-variable optimization problem. Depending
on
what optimizer you choose, all or some of the variables (in this case
3 rotations and three translations for a rigid transform only) might
be perturbed at a time.

2. Do you know some correspondence between the images a-priori? E..g.
do you know the location of at least 3 landmarks in both the images?
If yes, you can try aligning pointset to pointset (assuming the
bodies
they are on stay rigid). Search for a paper by Besl and McKay I
believe on google...on shape registration. If you dont know the exact
correspondence, you can still pick a finite number of landmarks
automatically from both images (but on the same object) and employ an
ICP like method to align two pointsets. I doubt if you will get sub-
pixel accuracy this way though.

3. If however you dont have this info, you can still try registering
the
images using statistics based methods. Look for 'mutual information
based image registration'. You can choose from among a number of
metrics depending upon the nature of images etc. For a good
combination of metric and optimizer, you might be able to hit sub-
pixel registration accuracy.

Here are some releated postings:

http://groups.google.com/group/sci.image.processing/browse_thread/thread/16b179374ca43da7/60c1099f4412333a?lnk=st&q=&rnum=19#60c1099f4412333a
http://groups.google.com/group/sci.image.processing/browse_thread/thread/449e1bdc9341bb20/2dc2872200b94a8b?lnk=st&q=&rnum=64#2dc2872200b94a8b

HTH,

P.

[Ivano Lopresti]

Pixel.to.life:

> I saw quite a few questions on how to estimate a rigid/affine

> transform between two given sets of images [...]

> do you know the location of at least 3 landmarks

> in both the images? [...]

In this case the problem of estimating a rigid transformation
can be reduced to the so called "orthogonal procrustes problem".
Maybe you'll find usefull the paper:
"Distribution of Target registration error in rigid-body point-based
registration" by J. M. Fitzpatrick and J. B.West.