your mail

[Peter Suetterlin]

  Hi Pablo,

welcome back :-)

> > For my old Powershot A50 I ended up doing a 'flatfield', i.e., taking
> > many images of clear sky (or white paper?  don't really remember).
> > Then you can see the intensity gradient over the field.  I wrote a
> > small C program that corrected this gradient, and it largely improved
> > the stitching results.

> what exactly did you do to correct the vignetting? I simply added the
> brighness difference I got from a white wall shot to the picture (using a
> layer).

Well, as I said I'm doing this 'out of' the standard image programs.
I am using IDL (from www.rsinc.com), an image processing language.  So
I added all the images up (as long integer), and then fitted a radial
polynomial to the intensity distribution (3rd degree, if I remember
correct - source code is at home...).  The polynomial is normalized to
1. in the center.

I read the images I want to correct and divide each channel by the
polynomial, then round them back to bytes.  higher values are clipped.
This gives errors if there are real bright features in the corners,
but I lived with that....

> Before addition, it was heavily smoothed (with a gauss filter), to
> remove the small structured details on the wall.

I tried that before, but it was still not satisfying.  Therefore the
fitting approach.

> I'm quite sure that a simple addition, like I did is the wrong
> way. Probably one should transform the picture to HSV and divide the
> V channel of though a brightness factor image.

True, as vignetting is (in principle) multiplicative.  For small
deviations you might set 1/(1-x) = 1+x, then addition is OK.  But that
should fail for real vignetting (fisheye lenses etc.)


> Hmm, probably this depends on the apperature, focal length and lots
> of other stuff.

Especially f-ratio.

> so it might be useful to try and shoot a reference images under similar 
> conditions (with the same exposure lock used for shooting the panorama
> pictures).

My old camera didn't have any f-setting, so I couldn't experiment with
that.  With the new one I did not see substantial vignetting so far.
Something to look at in long winter nights.... *g*

> this might also make the pano tools color correction work better,
> since these effects are strongest in the outer, overlapping regions,
> that are used by PTStitcher to do the color correction.

A loud and definitive 'yes' to this.

  Pit