[CREATE] Lens correction database

[Pablo dAngelo]

Hi Andrew,

Andrew Zabolotny wrote:

> > One problem with the "sheet calibration" approach is that it does
> > not reflect the typical shooting situation, where the objects are
> > (relatively) far away from the camera. The distortion is also focus
> > dependent, this effect is stronger if the object is very close to the
> > camera (think A4 calibration sheet). A general rule of the thumb is
> > to calibrate in a similar distance as the final range.
> Indeed, focal length slightly changes with the focus. That should be
> the reason why distortion slightly changes too.
> 
> For now lensfun assumes that distortion depends only on the focal
> length. I'm not sure how to handle this slight deviation of the focal
> length with focus. Either the calibration should be tied to the "true"
> focal distance, or calibrate by two factors - focal distance and
> distance to subject.

Actually, my suggestion would be to just ignore this issue, and calibrate
the lens at a distance > 1 m. A few months ago I started evaluate the
influence of the object distance on the distortion with several of my
lenses, but unfortunately got sidetracked by other stuff. If I remember
correctly, the influence was only really visible at very close object
distances.

> The first approach seems more correct, but it's a question how to
> compute the "true" focal distance. Maybe this deviation can be
> calibrated too. Then the "true" focal distance (and FoV) could be
> computed from the focal distance recorded in EXIF, and the
> distance-to-subject, which user will have to enter anyway (for
> vignetting correction).

Hmm, does the vignetting really change noticeably with
varying focus? I haven't made any analysis in that direction
yet.

> This approach has also the advantage that the
> current database won't have to be modified, just that it'll be a little
> inexact.

My feeling is that introducing such a focal distance vs. focus formula
or even calibration is not worth the extra effort. Especially given that
the EXIF focal length itself is usually not that accurate either.

I guess we should start with what we have now, and get it to a
stage where it can be actually used.

> If this will be done, it won't depend if you shoot the target at 1m or
> at infinity. But this rises the question how to calibrate the focal
> distance deviation dependence of the distance-to-subject.
> Making a 360 degrees panorama with a 300mm lens is a PITA :-)

The Zhang algorithm, for example as implemented in Bouguet's 
matlab toolbox or the CamChecker program are suitable for this,
and will likely provide higher quality estimates as a 360°deg
panorama.

> > However, it will definately better than no calibration at all. I
> > don't believe the vignetting can be recovered faithfully from it
> > though.
> For vignetting we have PTLens anyway ;-)

PTLens is not open source and when I last looked,
PTLens only offered an interactive vignetting correction.

> > Btw. after the most pressing issues with hugin have been fixed and a
> > new stable version is released, I'll look into adding an automatic
> > "submit calibration to lens database" feature.
> Sounds great. Do you mean to use the PTLens database, or lensfun?

The idea is to collect the data, possibly refine it and then include it in lensfun.

I have no connections to PTLens, and as you know, PTLens is a
closed, commercial product now.

Btw. how is progress on the ufraw integration?

ciao
  Pablo
_____________________________________________________________________
Der WEB.DE SmartSurfer hilft bis zu 70% Ihrer Onlinekosten zu sparen!
http://smartsurfer.web.de/?mc=100071&distributionid=000000000066