Camera correcting lens distortions?

"Hans-Georg Michna" wrote in message
...
[]
One way to increase picture quality could be to take several
photos in quick succession, then process the whole series,
concentrate on the one that is least blurred, take details of
moving objects from other shots in the sequence, etc.

Hans-Georg

Already partially available as "best shot selector" in the Nikon Coolpix
series. Selects the single "best" (i.e. sharpest) of a series of
images..... Doesn't do the image combination, though!

Cheers,
David

Hans-Georg Michna writes:

I don't have an answer to your question. Only wanted to remark
that any recalculation that involves pixel relocation will
necessarily reduce sharpness.

Not necessarily, at least not visibly. A high-quality resampling
algorithm (e.g. Lanczos-windowed sinc) will allow you to make small
local scale changes (which is all that's needed to correct moderate
barrel or pincushion distortion) with no visible loss of information.

I guess that this is one of the
reasons why nobody may be doing it in the camera.

Good resampling is expensive in CPU resources (lots of multiplies and
adds), which is more likely why nobody is doing it in the camera.

Dave

Ya! This one. http://www2.keck.hawaii.edu/geninfo/about.html

--
John
johnf 202 at hotmail dot com


"Matthias Heiler" wrote in message
...
| Hi group,
|
| Is there a camera that electronically corrects geometrical lens
| distortions (esp. Barrel)? For the fixed-lens digital cameras out
| there this should be not hard to implement. It also would
| manufacturers allow to use cheaper/lighter glass on their consumer
| products and still get decent image quality.
|
| On a similar note, does anybody know a decent lens-distortion
| correction software for Linux? It should be able to work on many
| images automatically, without manual intervention.
|
| Thanks,
|
| Matthias

Dave Martindale writes:

With a typical non-SLR, there's only one lens. You still have to model
its distortion as a function of the things you name, but a table could
be calculated automatically by the lens design software and included in
the firmware of the camera. Better yet, each camera could have an
individual table based on the measured performance of that specific lens
as it came from manufacturing, not the lens as it was designed.

It would be cheaper to just build a better lens.

--
Transpose hotmail and mxsmanic in my e-mail address to reach me directly.

"David J Taylor"
wrote in message ...
SNIP
.. I should have patented this idea - although I expect someone already
has.

Not exactly in camera yet, but looks interesting:
http://www.dpreview.com/news/0402/04...oopticspro.asp

Bart

"Mxsmanic" wrote in message
...
SNIP
It would be cheaper to just build a better lens.

Maybe, maybe not. A lens is ALWAYS a compromise between different
corrections. Improve one, another one gets worse. There are too many degrees
of freedom in lens design.

http://www.dpreview.com/news/0402/04...oopticspro.asp may somewhat
reduce most residual artifacts after the fact, although at the expense of
resampling which will reduce resolution a bit.

Bart

Matthias Heiler schrieb:

On a similar note, does anybody know a decent lens-distortion
correction software for Linux? It should be able to work on many
images automatically, without manual intervention.

Try hugin (hugin.sf.net) to find the parameters for your lens, and then
use them in a script with Panorama Tools.

Walter

In message ,
Hans-Georg Michna wrote:

I don't have an answer to your question. Only wanted to remark
that any recalculation that involves pixel relocation will
necessarily reduce sharpness. I guess that this is one of the
reasons why nobody may be doing it in the camera.

There's most likely some software though.

The best way, I would think, is to treat the three color planes of the
RAW image data as elastic bitmaps, so that software can work without the
restrictions of a raster grid, more or less like a plane object with
millions of vertices in a single plane, in a 3D program. The three
layers could be moved together for lens and perspective corrections, and
they could be moved independently of each other to counteract chromatic
abberation. Then, the data could be be rendered to a gridded raster
file of a size much larger than the original capture, without
significant loss or distortion of detail, for printing purposes, or
rendered at a smaller size for full image viewing onscreen, if printing
is not needed. The whole idea of moving detail from one pixel to
another is from the stone age.
--


John P Sheehy

In message ,
"David J Taylor"
wrote:

This sort of distortion correction is built into both Paint Shop Pro and
Pano Tools. Of necessity, there is a sharpness reduction. Whether
correction would be better done _before_ the Bayer interpolation or after
is something I would like to know.

Obviously, for the highest quality, the corrections should be done on
the RAW data, with total freedom of movement for the pixels, like a
plane object with millions of vertices in a 3D program, as part of the
RAW conversion process. RAW conversion software would need to be
re-written to deal with the elastic bitmap rather than the rigid ones
they work with now. Images would need to be rendered at a larger size
than the number of capture pixels, for best effect (for printing).
--


John P Sheehy

In article ,
Mxsmanic wrote:

Dave Martindale writes:

With a typical non-SLR, there's only one lens. You still have to model
its distortion as a function of the things you name, but a table could
be calculated automatically by the lens design software and included in
the firmware of the camera. Better yet, each camera could have an
individual table based on the measured performance of that specific lens
as it came from manufacturing, not the lens as it was designed.

It would be cheaper to just build a better lens.

Probably not, unless the total lens run was quite low.