| Re: lens distortion [message #36428] |
Fri, 19 September 2003 11:17  |
JD Smith
Messages: 850
Registered: December 1999
|
Senior Member |
|
|
On Tue, 16 Sep 2003 15:10:04 -0700, Ale wrote:
> I need to develop an IDL routine to correct lens distortion
> (barrel/pincushion) for our CT camera. Where could I find suggestions or
> examples on this topics ?
>
Not too hard to develop on your own. Have a look at:
http://www.path.unimelb.edu.au/~dersch/barrel/barrel.html
Near the bottom, he describes the quartic radial polynomial which
achieves barrel/pincushion correction. You can use just one of the
terms to get decent correction, e.g.:
r_src=(b*r_dest^2+(1-b))*r_dest
The idea is to transform the radius to each pixel using this equation,
remap to cartesian coordinates, and use INTERPOLATE to get the
results. Here's what I came up with:
function pincushion_correct,im,b
d=size(im,/dimensions)
;; Create, scale and correct a radius to each pixel
scale=.5*min(d)
r_dest=shift(dist(d[0],d[1]),d/2)/scale
r_src=(b*r_dest^2+(1.-b))*r_dest*scale
;; Convert back to cartesian coordinates
l=lindgen(d) & dy=d[1]/2-l/d[0] & dx=l mod d[0]-d[0]/2
angle=atan(double(dy),double(dx))
r_theta=[reform(angle,1,product(d)),reform(r_src,1,product(d ))]
new_xy=cv_coord(FROM_POLAR=r_theta,/TO_RECT)
x_src=reform(new_xy[0,*],d)+d[0]/2 & y_src=d[1]/2-reform(new_xy[1,*],d)
return,interpolate(im,x_src,y_src,/CUBIC)
end
And then try:
read_jpeg,filepath('people.jpg', SUBDIR=['examples','data']),im,/GRAYSCALE
tv,[im,pincushion_correct(im,-.15)]
Note that typical corrections for rectilinear lenses are around -.02
to .02 (this was exaggerated for demonstration purposes), and that
negative b's correct barrel, positive b's correct pincushion. This
same technique could be used for any distortions, including "rubber
sheet". Be aware, however, that IDL's native interpolator isn't the
best; see http://www.path.unimelb.edu.au/~dersch/interpolator/interpol ator.html
to find out how various interpolation algorithms stack up when used
with imaging data.
Good luck,
JD
|
|
|
|
|
|
|
|
| Re: lens distortion [message #36510 is a reply to message #36428] |
Mon, 22 September 2003 04:53  |
Ale
Messages: 6
Registered: April 2002
|
Junior Member |
|
|
I tried your routine but I received this error :
IDL> .COMPILE "C:\WINDOWS\Desktop\lente.pro"
% Compiled module: PINCUSHION_CORRECT.
IDL> read_jpeg,filepath('people.jpg',
SUBDIR=['examples','data']),im,/GRAYSCALE
% Compiled module: FILEPATH.
% Loaded DLM: JPEG.
IDL> tv,[im,pincushion_correct(im,-.15)]
% Compiled module: DIST.
% SHIFT: Expression must be a scalar in this context: <LONG Array[2]>.
% Execution halted at: PINCUSHION_CORRECT 6 C:\WINDOWS\Desktop\lente.pro
% $MAIN$
Perhaps I did something wrong.
Thank you for your help
Ale
"JD Smith" <jdsmith@as.arizona.edu> ha scritto nel messaggio
news:pan.2003.09.19.18.17.53.555731.30604@as.arizona.edu...
> On Tue, 16 Sep 2003 15:10:04 -0700, Ale wrote:
>
>> I need to develop an IDL routine to correct lens distortion
>> (barrel/pincushion) for our CT camera. Where could I find suggestions or
>> examples on this topics ?
>>
>
> Not too hard to develop on your own. Have a look at:
>
> http://www.path.unimelb.edu.au/~dersch/barrel/barrel.html
>
> Near the bottom, he describes the quartic radial polynomial which
> achieves barrel/pincushion correction. You can use just one of the
> terms to get decent correction, e.g.:
>
> r_src=(b*r_dest^2+(1-b))*r_dest
>
> The idea is to transform the radius to each pixel using this equation,
> remap to cartesian coordinates, and use INTERPOLATE to get the
> results. Here's what I came up with:
>
>
> function pincushion_correct,im,b
> d=size(im,/dimensions)
>
> ;; Create, scale and correct a radius to each pixel
> scale=.5*min(d)
> r_dest=shift(dist(d[0],d[1]),d/2)/scale
> r_src=(b*r_dest^2+(1.-b))*r_dest*scale
>
> ;; Convert back to cartesian coordinates
> l=lindgen(d) & dy=d[1]/2-l/d[0] & dx=l mod d[0]-d[0]/2
> angle=atan(double(dy),double(dx))
> r_theta=[reform(angle,1,product(d)),reform(r_src,1,product(d ))]
> new_xy=cv_coord(FROM_POLAR=r_theta,/TO_RECT)
> x_src=reform(new_xy[0,*],d)+d[0]/2 & y_src=d[1]/2-reform(new_xy[1,*],d)
> return,interpolate(im,x_src,y_src,/CUBIC)
> end
>
> And then try:
>
> read_jpeg,filepath('people.jpg', SUBDIR=['examples','data']),im,/GRAYSCALE
> tv,[im,pincushion_correct(im,-.15)]
>
> Note that typical corrections for rectilinear lenses are around -.02
> to .02 (this was exaggerated for demonstration purposes), and that
> negative b's correct barrel, positive b's correct pincushion. This
> same technique could be used for any distortions, including "rubber
> sheet". Be aware, however, that IDL's native interpolator isn't the
> best; see
http://www.path.unimelb.edu.au/~dersch/interpolator/interpol ator.html
> to find out how various interpolation algorithms stack up when used
> with imaging data.
>
> Good luck,
>
> JD
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
