[CREATE] Lens correction database

[Andrew Zabolotny]

From Sun, 09 Sep 2007 17:46:44 +0200
Pablo d'Angelo <pablo.dangelo at web.de> wrote:

> The lf_corrector_apply does not allow different transformations for
> the color channels (which is required for TCA correction).
Actually I have only one TCA models right now, but more can be added
(just like there are multiple lens distortion models) or maybe the
linear model may be modified to include fish-eye lenses too (adding
one higher-order term perhaps).

> TCA behaviour of most lenses is close to linear, but some specific
> ones (especially some fisheye lenses) exhibit a non-linear
> relationship. I provided Michel Thoby with some software to measure
> the TCA, and he has put a lot of work into testing different lenses
> and TCA correction tools.
Interesting article. How exact the TCA measurements were? Does Michel
realize that he actually measures a heavily post-processed image, e.g.
de-Bayerizing process can introduce quite significant errors (up to
a pixel) in color layers.

> To me it seems that photoshop is doing more than just a mere scaling
> of red and blue channels.
Maybe, never seen nor used it. Does it have a single control for
manual TCA correction or several?

> Also, TCA should not vary with different aperture settings. It will
> probably vary a little bit with the subject distance.
Hmm, interesting. How to explain then that many tests shows that
chromatic aberrations vary by changing aperture? For example
photozone.de always tests TCA at most aperture values, and there's a
visible trend usually.

> There is a nice paper (which I don't have handy right now :-( ) on
> how the k1 distortion coefficient can be interpolated across
> different focal lengths, by fitting a exponential function to the
> measured (distortion,focal length) values. This is much better than
> linear interpolation.
Cool. Right now I've implemented Catmull-Rom splines for linear
interpolation, I hope this is good enough, if there's no special
knowledge related specifically to optics.

> I have been doing some work on the stability
> and estimation procedure for the radial distortion part, but I need
> some more work to reach final conclusions. However, the traditional
> k1, k2 model seems to work quite well for most rectilinear lenses.
That's one task I have in mind too. After finishing the first version
of the library and integrating it into UFRaw I would like to implement
a tool that should make creating a lens profile a pretty easy task,
which will help in filling the database quickly. I have some ideas now
but perhaps it's too early to talk about it.

> I have written some example xml files and a tool to convert the
> PTLens database that might be helpful for inspiration, they are
> available at:
> http://hugin.svn.sourceforge.net/viewvc/hugin/lensdb/trunk/
Sure, I have started with your work. However, sine I changed the xml
format a little I wrote my own python converter; in any case it's a
quite temporary tool since it won't be needed in long-term.

-- 
Andrew
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://lists.freedesktop.org/archives/create/attachments/20070910/b5969865/attachment.pgp