Using hugin / panotools to help blend exposures.

[Pablo d'Angelo]

On Sat, 25 Oct 2003, Ed Halley wrote:

Hi Ed,

nice. We're getting lots of interesting posts here lately :)

> This morning, when digging back through archives for a nice panoramic
> series to play with Hugin, I realized that Hugin and panotools could be
> used to assist in aligning two images that were hand-held.
> 
> Here's the results of my first experiment in using panotools to help
> mix/merge two exposures for dynamic range from handheld bracketing.
> 
>    http://www.halley.cc/pix/mixing.with.hugin.jpg
>    http://www.halley.cc/pix/mixing.with.gimp.jpg
>    http://www.halley.cc/pix/mixing.results.jpg
>    (May be removed in a couple weeks.)
>    Contact me if you want full-sized copies.

Looks good! But you could try to use better lens distortion values, the
sidewalk looks slightly bent. This might also improve that matching of
the output images.

> One, there's still no good way to get masked layers for GIMP.  I knew

Yep, thats a open problem. Either use http://www.vierpi.de/tif2xcf.html
but that needs two runs of PTStitcher.

Or you can try the patched tiff reader from Kai-Uwe, that should support
reading the PTStitcher "TIFF_m" (called "multiple tiff" in hugin) output.

Unfortunately I didn't have the time to try it out myself yet.

The code for the gimp tiff plugin has been posted to the ptx mailing list a
few weeks ago.

> that today.  If someone has the panotools source code, they could add a
> "save as separate .tiff RGBA files" (which would just name the
> temporaries and quit cleanly).  Saving as separate .png RGBA files would
> be a plus.

nobody has that code, so you're out of luck here...

> The goal of panotools is to align images along a fairly narrow edge,
> where feathering would transition from one image to the other.  However,
> even when both images were at the same lens, same focal length, same
> perspective (within an inch or so), the image scales were hard for
> panotools to rectify exactly.  I experimented with different control
> pairs, but could never get the distortions quite exactly to match.  I
> eventually ended up shrinking the underexposed frame by a factor of
> 0.9975, which got nearly pixel-perfect alignment over the whole image. 
> Nearly.

Hmm, so you have optimized the lens distortion parameters already?
Having good values for a,b,c,d,e is really the key to a pixel accurate
registration. (and not changing the viewpoint.. not sure how big the impact
of a 1 inch camera movement is on your pictures. But even with handheld
shots you can easily get below an inch displacement between shots.)

ciao
  Pablo
--
http://wurm.wohnheim.uni-ulm.de/~redman/
Please use PGP