annala@neuro.usc.edu (A J Annala) (12/10/88)
I would appreciate receiving information, pseudo code algorithms, and/or
actual source code that would enable me to assemble a single large image
from a series of overlapping images (512 X 384 pixel) acquired from NTSC
video rate 8 bit CCD camera. Ideally, a least squares minimization code
should enable me to accomplish the registration process. However, I am
open to examining any other approach to solving this problem.
It seems to me that astronomers working on assembling mosaic images of
the moon, other planets, and their satellites should have this problem
pretty well solved. In addition, people assembling LANDSAT and/or SPOT
earth resources satellite images should have workable code.
My images are even simpler than these space based applications. I will
be acquiring overlapping NTSC video frames from brain sections examined
in a Zeiss Axiophiot light microscope. The images will be preprocessed
to remove artifacts induced by nonuniformities in illumination of the
microscope field, dark current, etc. I will acquire the images normal
to the plane of section. No orbit skewing, sun shadow angle, or other
similar distortion should be present in the starting images. However,
the least squares (or other registration algorithm) will have to work
with roughly 1.0 OD unit dynamic range continuous grey scale images to
accomplish the registration process.
Please email responses to:
annala%neuro.usc.edu@oberon.usc.edu
Thanks, AJ Annala, USC NeuroScience Program