UBC computer science PhD candidate Matthew Brown, 25, has
developed panorama software with a new object recognition
feature that surpasses the capability of panorama-building
software currently on the market.
Brown’s AutoStitch Panorama software can automatically
recognize and match images that are similar. The software
then “stitches” the images together to create
a seamless panoramic view of up to 360 degrees. All the user
has to do is download their digital photos from their camera.
The software’s ability to automatically recognize unordered
image sets represents a major step forward in object recognition
and computer vision, says Brown’s supervisor Professor
David Lowe, a leading researcher in the field.
With currently available Panorama software, photos have to
be carefully taken in a fixed sequence, downloaded and then
manually identified and aligned by the computer user. It’s
a process that takes time and some technical expertise.
With Autostitch Panorama, the matching process is fully automated
– and quick. A standard PC takes about three minutes
to match and register all images and then render the panorama.
Brown, a native of Manchester, England, is hoping to improve
on that time in the future.
Brown and Lowe built on Lowe’s previous research to create
the Autostitch software, which uses a probabilistic model
to detect and verify similarities to match the images and
then automatically stitch them into the panoramic view.
Brown is set to present a paper on the research, entitled
Recognizing Panoramas, for the first time at the 10th International
Conference on Computer Vision in Nice, France, October 13-16.
Sample images produced by AutoStitch Panorama software can
be viewed at www.cs.ubc.ca/~mbrown/panorama/panorama.html.
As with the Panorama software currently on the market, the
final composite image is a computer graphic that allows the
user to explore the panorama by simply dragging the mouse
around the image. Images can also be mapped to the surface
of a sphere or cylinder to provide a 360-degree photograph.
While no special camera is required, one restriction the researchers
hope to overcome in the near future is to enable the software
to match images of one scene taken from a multitude of locations.
Currently a photographer can’t move around snapping
photos from multiple locations and use this software. If the
pair master that problem, they will have achieved something
that 20 years of research in this area has yet to conquer.
For now, they’re hoping an outside company will licence the
software and develop it further for commercial use. Virtual
tourism websites and online walkthroughs of interiors to sell
real estate are just two practical applications where Panorama
software is already in use.
After Brown returns from the Nice conference, he will be
heading to Microsoft and a four-month internship with Rick
Szeliski, a pioneer in this area of computer science research.
The UBC Department of Computer Science is a dynamic,
youthful, and growing community renowned internationally for
its excellence and depth of research. Recognized for teaching
innovation, the Department places a conscious focus on interdisciplinary
programs. There are approximately 900 undergraduates, 185
graduate students and 41 full time faculty.
-30-
