proj-imim:
Re: Pixel Picker questions
Peter Murphy 2000-May-09 22:58:20
Some thoughts on recent point picking questions -
it is very important to scan negatives the same way each time -
ie with the same cropping and symmetrically with respect to the
taking lens. I can get really small discrepancies (effectively
perfect) with digital images but often not with scanned ones.
With fisheye film shots you can get good scanning correspondences
by using a circular path for successive images to define the selection.
Half a dozen points per image is fine and 3 or 4 is what I usually
use. A graphics tablet and a large monitor is very handy.
You dont need the same number of points per image.
Put the points spread over the overlap if possible but not
all near the edge. Even if your point discrepancies are only
2 or 3 pixels with say a 3200 pixel wide panorama this is
going to give visible detail breaks in some parts of the
overlap with many kinds of images and this is why final
tweaking via layers or hand stitching is what I usually do.
I am sure with a tripod shot/digital camera in low flare
situations or with
smaller pano sizes the stitching can be totally automatic
(after point marking). Flare and lens illumination fall off
mean that perfect results visually apart from point correspondences
are hard to achieve automatically. ImageStitcher (expensive)
from Realvz does really great stitching of large partial panorama
multirow
nonfisheye shots with great color/brightness blending accuracy
but for full panoramas/fisheye/mixed lenses/handstitching
Panorama Tools is the total champion.
Because of no Pixel Picker for PC I use Sascha Kerschhofers
'PTools PixelPicker' which just uses Javascript and your
web browser instead.
http://stud1.tuwien.ac.at/~e9127005/ptools/
This is just great and very convenient
for deleting a few points and adding a few more. You would
need to modify its script output a little to get your various
PTStitcher output options but for hand stitching
especially it is fine. It takes me with this about 10 minutes
to mark maybe 40 points with 8 images.
If I take and scan my pictures with any sort of care
I find that the Optimizer with Panorama Tools works to
at least 3 or 4 pixel accuracy with 3200 pixel panoramas
99 percent of the time - digital images are always
better. btw Helmut please some stuff soon
on PTInterpolate and PTMorph?
And PC versions??
Peter Murphy
The Swamp Gang wrote:
> The scary part is this is starting to make some sense. However, I do
> have a few quick questions for the group: Is there a best method for
> picking points to use? Should the points be samples from all over the
> overlapped areas or is it best to confine them to a specific
> region? Are there areas to avoid such as close to the edges? Does a
> higher number of points create a better quality pano? Do you have to
> have an equal number of points for each image; like n0-n1 6 points,
> n1-n2 6 points, etc or can one of the overlap areas have a different
> number than the other overlapped areas? I find this question comes up
> mostly when taking panos on the water where the sky and water on one
> of the overlapped areas don't give too many points to pick
> from. Thanks, Mike Yother