🚀 go-pugleaf

 tirone@neurocog.lrdc.pitt.edu (Steve Tirone) writes:
>
>	I've got several MRI images of the brain, taken over period of
>time (minutes), and there's some slight movement of the head between
>images.  I'd like to "register" these images, that is, align them so
>the images all overlap exactly.  I do not have any external markers to
>use as reference points (fiducials), BTW.
>
>	What is a good way to go about this?   It appears that without fiducials,
>automatic image registration has been attempted in various ways, from
>superposition of centroids and principal axes (?) to correlation using FFTs to
>contour matching via Hough transforms...anyone care to comment on what might be
>the "best" solution for my purposes?
>
>	Also, I'm wondering what kind of error (in the pixel values) I might
>introduce by registration and how I might quantify it.  Is it possible to align
>images with subpixel accuracy?   How does resampling the image affect my
>interpretation of the pixel value? (Like "its an average signal value, plus x
>percent error introduced by resampling after rotation" ??)
>
>	I'm a newcomer to image processing.  Any pointers, solutions, etc.
>very much appreciated.
>
>Steve
>tirone@neurocog.lrdc.pitt.edu


A big issue is whether the registration algorithm expects the kind of motion
present inthe image.  Some algs were designed to handle only 2-D
translations & rotations, not rotations in depth, or volume-changes.
Its an open problem.

--
David "Dances with Motion Sensors" Honig

"Don't you believe half of what you see, and none of what you hear."
	---Lou Reed