Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!cs.utexas.edu!uunet!seismo!beno!black
From: black@beno.CSS.GOV (Mike Black)
Newsgroups: comp.graphics
Subject: Re: Image restoration
Keywords: image restore warp
Message-ID: <49009@seismo.CSS.GOV>
Date: 21 Jul 90 13:43:53 GMT
References: <4187@tahoe.unr.edu> <1990Jul15.061617.19425@comspec.uucp> <48999@seismo.CSS.GOV> <1990Jul18.162340.12905@imax.com>
Sender: usenet@seismo.CSS.GOV
Distribution: usa
Organization: Center for Seismic Studies, Arlington, VA
Lines: 41

In article <1990Jul18.162340.12905@imax.com> dave@imax.com (Dave Martindale) writes:
>In article <48999@seismo.CSS.GOV> black@beno.CSS.GOV (Mike Black) writes:
>>I'm in need of an algorithm to undo the effect of a 90 degree field of view
>>lens.  I tried a dual-axis third order polynomial fit and it didn't handle
>>the corners very well.
>
>First, you have to know the distortion characteristics of the lens.
>Designers *try* to build lenses with minimal distortion, even if the
>field of view is 90 degrees, so the distortion that remains is one
>of the compromises that had to be accepted to get something else in
>the design.  Thus, every design of a wide-angle lens will have
>different distortion characteristics.
>
>Or was this lens designed as a fisheye?
>
>Anyway, once you know what the distortion you are trying to correct
>looks like, you can see how well various polynomials (or other functions)
>will fit the curve.

I maintain that you don't need to know a thing about the lens.  I've taken
an evenly spaced grid and pulled off x,y locations of where the points are.
I've found that a 3rd order polynomial fits quite well in both axis.  I've
got two problems now:
1.  I've found that the centroid of the fit changes based on which vertical
or horizontal scan line I references (it looks like another parabolic
function).
2.  I'm still a little fuzzy on how to figure out that:
	x',y' = f(x,y)
Using the target image coordinates.  In other words, I want to iterate from
x=0 to 511, y=0 to 511 and figure out where the distorted pixel is and put
it in a non-distorted buffer.  This will assure image fill.  I figure that
I will have to change to polar coordinates and then back again.  I'm having
problems finding out where the center of the field of view is so I can
compute the polar coordinates from that point.  I need to be extremely 
accurate and need to be able to prove the method.

--
-------------------------------------------------------------------------------
: usenet: black@beno.CSS.GOV   :  land line: 407-494-5853  : I want a computer:
: real home: Melbourne, FL     :  home line: 407-242-8619  : that does it all!:
-------------------------------------------------------------------------------