[ptx] Hugin : parameters for correction of chromatic aberration

[Pablo d'Angelo]

YMI2 schrieb:
> Hello,
> 
> In the panoramatools gimp plugin, when I want to correct  an image for
> transversal chromatic aberration,  I used radial shift correction with
> these parameters (corresponding to my lens) :
> r_src = axr_dest^4  + bxr_dest^3 + cx r_dest^2 + dxr_dest
>             a        b   c    d
> red :     0.005  -0.022  0  1.0174
> green :   0.005  -0.022  0  1.017
> blue :    0.005  -0.022  0  1.0174
> 
> I want to use Hugin to do the same correction on 16-bit tiff images. 
>  -> it's ok for the a, b and c parameters, but what about the d one ?
> Does it correlate with the degre of view v parameter ? If so, how ?

Yes it does. However, for normal rectilinear lenses, the relationship 
between v and d is not linear. It's probably the best idea to use a 
equirectangular input and output projection, since the relationship between 
image width and field of view is linear.

It should be possible to just use the d parameters as v parameters of the 
input images (single channels).

>  -> there seem to be no way to correct each color with different
> parameters as in the panotool plugin. Is it planned to include this for
> later release ?

Its planned for a later release...

 > For now, I must first decompose in 3 gray images, do the
> correction separately in hugin and then recompose. 

Erik Krause and Jim Watters have done some experiments in the same direction
See http://www.erik-krause.de/index.htm?./ca/index.htm for more details.
He also has a script that uses imagemagick to split and combine the images.

Some time ago I have had some success in splitting a Peleng 8mm fisheye 
image into 3 gray images, setting lots of control points, and optimizing 
them, to gain the required distortion parameters directly. It is a bit 
tricky, and a good image, with high contrast (not overexposed!) and good 
edges is needed.
Winter is a good time for such subject. lots of white snow with high 
contrast to roads, benches, trees etc.

Erik mentined that a similar approach failed when he tried, but it worked 
for me. I guess the subpixel accurate fine tune was really helpful there.

ciao
   Pablo