Vignetting: Correction with fulla
Lars O. Grobe
I am trying to use hugin and its tools do correct images taken with a
fisheye lense for distortion and vignetting. I do not create panoramas,
but I need accurate corrections as I will assemble the pictures to HDRs
and read out pixel values later to get numerical values.
I am curious about the vignetting correction. I would expect a curve of
correction values to be applied to an image, but I could not find much
information about how the correction parameters are found and if there
is a simple linear correction applied (which would not be enough for me)
or a non-linear correction depending on the distance to center is used.
Is there anyone knowing the details of this correction and / or anyone
who has ever used fulla in a way that numerical accuracy was of
importance?
TIA&CU Lars.
Pablo d'Angelo
Lars O. Grobe wrote:
> Hi group,
>
> Is there anyone knowing the details of this correction and / or anyone
> who has ever used fulla in a way that numerical accuracy was of
> importance?
For all detail see http://hugin.sourceforge.net/tech/
I haven't used it for physically accurate reconstruction yet, and I have no
experience how accurate it is. It should be quite good, but a proper
calibration setup is essential for such a task. It also heavily depends on
your application and equipment and whether there are other effects that need
to be compensated (lens flare etc.).
Can you tell us a bit more about the desired application, its accuracy
requirements etc?
You should use linear image derived from RAW (with dcraw, for example).
For calibration:
1. Using a Panorama: Use a scene with little contrast and far away objects
(avoid lens flares and parallax errors) and at least three images, best
arranged in a triangle.
To check if the correction is good, a independent measurement of the
vignetting could be performed by photographing a small, uniform target with
different camera rotations (be sure to respect the nodal point, to minimise
influences by changing viewing directions). THe plot the brightness of the
targets over the distance from the image center to see the vignetting curve.
That can be used to derive the correction polynomial and to check the
accuracy of the calibration done with the panorama.
I'm very interested in such a comparison, and wanted to do it myself, but my
fisheye (peleng) suffers from uneven color casts and heavy lens flare, and
is thus unsuitable for such an accuracy assessment.
ciao
Pablo
Lars O. Grobe
thank you for the reply! I see that I should go and find an environment
with distances and uniform lighting conditions now. For an almost 180
degree lense, this is not easy to find, trees standing around
everywhere :-)
I want to read luminance values from the HDR image that will be
generated from the corrected images. Accuracy is important over the full
field of view, I would like to stay <10%. I was thinking about using a
calibration with a known light source moving through the image first
(this has been done before and is documented), but than the idea came to
use fulla, as it supports both barrel AND vignetting.
Thanks, CU Lars.
Pablo d'Angelo
> Hi Pablo,
>
> thank you for the reply! I see that I should go and find an environment
> with distances and uniform lighting conditions now. For an almost 180
> degree lense, this is not easy to find, trees standing around
> everywhere :-)
The best bet would be a on top of a high building on an overcast day. But it
is probably not essential. Notice that the panorama needs to fit well
(output the remapped images and view them in difference mode in gimp or
photoshop).
> I want to read luminance values from the HDR image that will be
> generated from the corrected images. Accuracy is important over the full
> field of view, I would like to stay <10%. I was thinking about using a
> calibration with a known light source moving through the image first
> (this has been done before and is documented), but than the idea came to
> use fulla, as it supports both barrel AND vignetting.
I think less then 10% shouldn't be a problem. I'd still use your older
method of manual calibration to check. The idea is not to take a single
flatfield image (creating a 180° uniform target is indeed a challenge), but
to rotate the camera so that a image the same, small reference at multiple
placed of the image. Then the values measured at these points can be used to
derive the correction polynomial. Use a panorama head to rotate around the
no parallax point (aka nodal point).
ciao
Pablo