[ptx] Stitching scanned, "flat" images

[Rik Littlefield]

Philippe Gac wrote:

>hi all,
>
>Stitching scanned, "flat" images
>  
>
>>The proper way to do this with PT is to use the offset variables keeping
>>y,p,r constant just optimize d & e
>>    
>>
>and y, p, r should be set to a very small value
>OK
>Now, what about translated views for rectilinear projection of a front plane?
>you have to set y, p, r to their real values
>can the offset variables d and e do the job correctly?
>I'm not sure this can be done even if the picture is taken orthogonally to the plan
>
This should work OK but only if a/b/c are all zero.  If a/b/c are not 
zero, you will get distorted images for large d and e.  This is because 
d and e shift the lens center, which shifts the center of the lens 
correction polynomial.  If you require nonzero a/b/c to correct lens 
distortions, then you need to make two passes.  First correct the lens 
distortions by generating new images with y/p/r=0 and a/b/c/d/e as 
needed.  (Probably d/e=0 in this case.)  Then make a second pass to 
stitch the corrected images using d/e to do the shifts for alignment.

>What if the pictures are taken with vaious angles to the plan?
>this happens systematically because of the imprecise positioning and intentionnally when you want to avoid the flash
>reflection on the plane.
>
Sorry, won't work, or at least won't work very well.  The Panorama Tools 
imaging model assumes that the world is painted on the inside of a 
sphere, with the lens entrance pupil at its center.  The math for this 
model also handles a few other special cases.  One of those special 
cases is translation in a plane, perpendicular to the line of sight, 
parallel to a flat subject, using a perfect rectilinear lens.  Most 
everything else is an approximation and you will get stitching errors.

--Rik
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.email-lists.org/pipermail/ptx/attachments/20041009/bfcfa7a9/attachment.htm