[argyllcms] Re: scanin and perspective distortion
- From: "Guy K. Kloss" <G.Kloss@xxxxxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Fri, 15 Feb 2008 16:05:40 +1300
Hi,
On Thu, 14 Feb 2008 2:09:26 am Graeme Gill wrote:
> After thinking about this a bit more, I decided that correction
> for perspective distortion could be added to scanin without major
> changes to the architecture.
Wow. Didn't expect that, but that would be awesome for everyone wanting to use
Argyll with digital photography or any other cameras.
> It may be possible to add barrel/pin cushion correction at the
> optimization stage, but I'm not convinced this is needed in
> many cases, unless the distortion was severe (fish eye lens perhaps ?).
All depends on the amount of distortion. Even with normal lenses/focal
lengths, if the image of the target is pretty much format filling, it can be
of a problem if many, small patches are present. Then the necessary shrink
reduction for proper patch detection would massively decrease the patches'
real estate usable for measurement.
Of course, one may be able to take pictures where the target occupies a lower
angular field of the image to yield lesser distortions. This would also
reduce the effect of vignetting. However, for low resolution cameras or
constrained geometries this might be not an option. This is also the case I
am facing here, but I'm pretty sure I will find a way around that ... The
most important is to be able to reduce glare by angling the target, and
therefore can still use the result as perspective distortion can be
compensated.
> Vignetting is another story. I imagine this is best taken
> care of at the patch value level. The ColorChecker charts
> both have a ring of black/grey/white shares around
> their periphery and in the centre, and I'd imagine
> this is to allow some level of correction for
> uneven lighting and/or Vignetting.
This leads me to a problem I have just encountered with the Wolf Faust C1
target: Due to the organised patch arrangement (in contrast to some sort of
randomised ordering) I am getting distinct colour distortions in certain
areas. E. g. with glaring reflects in the middle of the chart, I got some
strong distortions in the cyan areas, as all cyan patches are co-located in
one column.
> Maybe some approximate
> correction could be achieved by fitting a surface to the
> black/grey/white planes (perhaps by setting up a rspl for
> each patch), and then fitting a low order curve the three
> values at each patch, and then computing a correction. It may
> be that doing this in the device space is sufficient to
> improve the accuracy, or something trickier involving
> creating a transfer curve using the middle set of
> grey wedge patches may be the way to go.
Very good idea.
Guy
--
Guy K. Kloss
Institute of Information and Mathematical Sciences
Te Kura Putaiao o Mohiohio me Pangarau
Room 2.63, Quad Block A Building
Massey University, Auckland, Albany
Private Bag 102 904, North Shore Mail Centre
voice: +64 9 414-0800 ext. 9585 fax: +64 9 441-8181
eMail: G.Kloss@xxxxxxxxxxxx http://iims.massey.ac.nz
Other related posts:
