On Tue, 27 Feb 1996 10:14:02 -0500 Charles Richardson wrote:
 
 
>
> IDEALLY, I would like to hold one slice image on the screen, then
bring
> the next one in the stack up on the screen to overlay the first
image, then use
> the mouse to rotate and translate, then "lock" it down when
aligned...and so on.
> Probably a pipe dream.
>
 
>
> Has anyone else solved the problem of re-registration of histologic
specimens?
>
 
 
I have run into a similar problem. The question is, how constant are
reference points between slides? If you process the image to get a
clean "outline" you can then flood fill inside that outline to create
a solid reference image. You then use an iterative procedure to
automatically do alignment. First you cross correlate each pair
of rerence images to calculate their x-y offsets. One of  the pair
is moved by this offset and then rotated a bit. A new cross
correlation is performed. If the correlation is stronger then you
should be moving/rotating in the right direction. This rotation/cross
correlation is repeated until the cross correlation is maximised.
By summing all the rotations and translations you have the
corrction to apply to the original images. A program such as IDL can
do this sort of thing very well. The trouble is that this algorithm is
not robust unless the changes between each pair of referenec
images are small... In orther words, if the sections are distorted by
mounting then aligning them is very difficult. If you have reference
points in each slide that correspond to reference points elsewhere
then you can warp the images to align reference points.
Basically, if its hard to do by eye then doing it automatically is
very very difficult.
 
Hope this helps -let me know if you come across a better method.
 
Regards
 
Mark Cannell
email [log in to unmask]