Hugin can now correct the panorama for exposure, vignetting, white balance and
response curve. The combination of geometric and photometric alginment
also enables the creation of HDR panoramas directly in the process.
The algorithm can be used in fully automatic mode. It took 7 seconds to
estimate vignetting and camera response of the 61 image Venice panorama shown
below.
The method is described in this panorama stitching paper. The presentation slides are also available.
Some Preliminary Results
These results can be compared with previous work by Dan Goldman and
Jing-Hung Chen. Compared to their algorithm, the method in hugin can
properly handle white balance differences and is a lot faster, because it
does not need to solve for the scene irradiance directly. The implementation
in hugin will also allow the creation of extended and high dynamic range
mosaics
Green Lake Panorama
Goldman and Chen have used this image sequence in their paper. Their results
can be directly compared against the images shown here. The images were captured
in aperture priority mode. By examining the source images, small white balance
differences between adjacent images can be detected, especially at the right border
of the panorama.
Venice anorama
This panorama consists of 61 image, captured by Jeffrey Martin,
using
a Canon 5D with a manual focus Yashica 300mm lens. The images have been
captured in manual mode to ensure constant exposure and white balance.
As seen in (a), this lens suffers from very strong vignetting. When blending
the images with enblend (using 20 blending levels, to ensure a blended area),
large, unpleasant residuals of the vignetting remain, especially in the sky.
After correcting the vignetting (b), a very good result is obtained by just
just stacking the images without any seaming. The remaining small defects can
the easily be removed by enblend.
MetaPanorama by Gadl
This equirectangular panorama consisting of 60 images has been shot by Gadl
with a Sony DSC-T5 camera which only provides an automatic exposure mode. Thus
the images have been captured with varying exposure and aperture settings (f3.5
and f5.6). The visible seams in the sky are due to moving clouds, and not
caused by photometric misregistration. This scene also shows that the proposed
method can robustly handle images with some moving objects - such as clouds -and works
on large, equirectangular mosaics.
Gadl has managed to do a very good job with assembling and
post-processing the panorama, using only enblend and PTblender. However the
new method is much simpler to use.
When comparing the images, please keep in mind that the images below have
not been processed for tone curve and color in GIMP or Photoshop, they are the
raw results, straight after stitching with hugin. This example also shows that
enblend is quite good in compressing the large dynamic range, since each
picture predominatly contains one useful exposure setting. By close inspection
of the panorama produced by enblend, one can see that some areas, such as the
two other photographers and the tree with the bicycle suffer from strong
blending artefacts due to large exposure differences. Enblend 3.0 has been used
without the new seam-optimisation (it crashed when it was enabled), and the
images have been blended in exactly the same order
Click to see full size image
a) Aligned Originals - No blending
Click to see full size image
b) No correction - Only blending
Click to see full size image
c) Vignetting, exposure and white balance correction - No blending
Click to see full size image
d) Vignetting, exposure, white balance correction - Blending |
