|
|
Thursday, October 2, 2003 |
A new stitching
system: Matthew Brown, a PhD student at UBC in Vancouver,
Canada, recently
wrote the first system to recognise and stitch panoramas fully
automatically from an unordered set of images. It uses a new class of
generic image features and object recognition techniques to
automatically detect matching images in an image set. It then
automatically registers and stitches each panorama in the image
set that it finds. The results of this are online.
Also available at the same site is a paper entitled "Recognising
Panoramas" by Matthew Brown and D. G. Lowe, which is being published
at the International Conference on Computer Vision this year.
From the webpage: "This allows photographers to input to the system the
images on an
entire flash card or film, and the system will automatically recognise
and stitch panoramas without user input." The software can extract
panoramas of differing orientation from a batch; i.e. some of the
panoramas in the batch can be portrait, others landscape.
The software is called AutoStitch, and can handle full 360x180 degrees
panoramas, as well as cylindrical panoramas. It runs on Linux and is available on license
from the University of British Columbia, and the "SIFT" feature matching is patented.
The AutoStitch software sounds quite fast at what it does, yet it is a
research implementation. It currently takes about 2 minutes to
automatically match, stitch and blend panoramas from 50 input
images.
My opinion: this is a
fantastic advance in panoramic photography.
The time savings alone make it an exciting development. Imagine no
longer having to tediously copy each panorama's images into unique
directories, tell the stitcher what kind of lens you used (if you
can remember), then drag each image one at a time into some sort of
stitcher "waiting room" panel, and then drag those images into position
in the stitcher itself, then tell the stitcher to stitch each pair or
set together. Instead, this software can look through a batch of images
downloaded from a CF card and figure out which images belong in
panoramas, infer the lens, and
stitch them together. From looking at the sample images, it appears to
do a very good job of stitching and blending. The amount of time saved
by such a process is significant--I would guess it could save me 4-8
hours of mind-numbing computer tedium per day trip's worth of
panoramas.
This is the type of
advance I've been calling for in my
essay.
I would guess that all stitching applications would want to have
the auto-matching feature. I'll post more information as I get it.
7:00:00 AM
|
|
© Copyright 2006 erik goetze.
|
|
|
|
|
Purpose |
VRlog provides news, developments and analysis of the virtual reality (VR) world from a nature photographer's perspective. Since I am not connected to or funded by any VR vendor, I intend to objectively appraise what's going on, and the direction VR is headed in. -- erik goetze
|
|
|
|
Syndicate VRlog |
The items on
this site are available in an RSS newsfeed, an XML file format.
|
|
|