In March 2011 I tried in vain to align 3
images with Hugin so that I could
compare them with mouseover image manipulation (run the mouse
cursor over the images below to see what this means).. The main problem was that one
of the images was taken at a different focal length, so they didn't line up. I asked on the
Hugin mailing list, but
didn't get the answers I wanted. In February 2013 I finally worked it out—I thought. Since
then I've discovered a number of cases where this method didn't work, and I'm still trying
to work out when it does and when it doesn't. Read on and try it; feedback more than
welcome.
Here's what you need to do, based on the images I took 2 years before. As you can see,
they're not perfectly aligned with each other. Further down are versions which are
correctly aligned. Run the mouse cursor over the images to compare with the following
image, or click on the images for larger versions:
In this article, all images are shown in relatively small format. Click on an image up to 3
times to enlarge it.
The steps are:
Load the images into Hugin, of course.
In the Photos tab, select Create control points. Depending on the
images, CPfind or panomatic is better. In this particular
example, CPFind did not find any control points between the middle image and the
other two.
If this works, as above, then all is well. But it seems that if the focal lengths are
too close together, the optimizer fails, and you get a message like:
If this happens, you can try selecting Positions (y, p, r), but in my experience
this doesn't often work. I don't have a solution for this case.
Another case where this doesn't work is if the images have been taken from a slightly
different position. In this case, the selection Positions and Translation
(y,p,r,x,y,z) might work. In general there's room for experimentation with various
optimization parameters.
In the first case, the images now line up well. If necessary, return to the Control
Points tab and edit the worst control points.
In the Fast Panorama Preview window, all the images will be on top of each other, and
probably far too small:
Move to the Stitcher tab. Select Calculate Optimal Size,
deselect Panorama Outputs, and select No exposure correction, low dynamic
range in the Remapped Images section, then Stitch!.
=== grog@eureka (/dev/pts/7) ~/Photos/20110330 55 -> ls -l example* -rw-r--r-- 1 grog lemis 25264578 10 Feb 12:14 example_exposure_layers_0000.tif
-rw-r--r-- 1 grog lemis 24988642 10 Feb 12:14 example_exposure_layers_0001.tif
-rw-r--r-- 1 grog lemis 25387168 10 Feb 12:14 example_exposure_layers_0002.tif
Here the results, rearranged in time sequence. Run the mouse
cursor over the images to compare with the following image, or click on the images for
larger versions:
If you want to try this with these images, they're here, here and here. In total they're about
8 MB.
In passing, it's worth noting how little you can rely on the information in
the EXIF data. According to that data, the
first image was taken at a focal length of 23.0 mm, and the other two were taken at a
focal length of 21.0 mm. Hugin reports focal lengths of 21.98972 and
21.45713. While these may not be absolutely accurate, the difference in focal length
is probably close enough—only 0.5 mm instead of the 2 mm that the EXIF data suggests.
