Hi Andrew,
Thanks for the links. I was aware of ZoneTags. I believe a lot of it
came out of Mor Naaman's Phd and subsequent research - which is really
great work. We are also looking at different resources for descriptors
including web sites like Flickr and Wikipedia. I guess the main
difference is that the Flickr tags tend to give toponyms (e.g. "Golden
Gate Bridge"), which are definitely important to us, but we also want to
be able to index images using more generic geographic concepts e.g.
"tropical beach" - which is where this work comes in.
You're right too about the issue of only having Lat, Long coordinates.
We have different goals that were aiming at which relate to different
levels of richness in spatial metadata, going from Lat, Long, and
altitude right the way to azimuth, pitch, and roll. We're investigating
cameras that can capture those parameters at the moment. We will also
using other information, for instance we have a partner who works in
content-based analysis of images, which should allow us to discriminate
amongst some of the examples you list. Also just by looking at the other
Exif metadata e.g. zoom and focal length should help us to identify
particular types of shots, like close-ups, that we know we can't deal
with. Actually, I think it will be very interesting to see the extent to
which location really isn't enough when its combined with other sources
of information.
Cheers Alistair
Andrew Turner wrote:
On 8/1/07, Alistair Edwardes <[EMAIL PROTECTED]> wrote:
The project is called Tripod (www.ProjectTripod.org). It's based on the
premise that the geography of where a photograph was taken plays a
central role in how it is described. Our aim is then to automatically
caption images using high level descriptors that have been derived from
analysis of geographic databases at the location where the photo was
taken. We'll then use more conventional technologies from information
retrieval to index these and hopefully allow them to be better searched
using textual queries.
Hi Alistair,
you may also be interested in checking out the Yahoo Research's ZoneTag project:
http://zonetag.research.yahoo.com/
and their TagMap Project:
http://tagmaps.research.yahoo.com/
They have been pulling the tags from geolocated photographs in all of
Flickr (not a small database) to build these kinds of databases like
you suggest. The ZoneTag API lets you easily query this database.
I've built a Firefox GreaseMonkey script that you can use on any of
your photos in Flickr to pull up suggested tags based on ZoneTag.
But as your project seems to be addressing is that just given a
location isn't enough. If I'm standing on the south bank of the Thames
in London I could be photographing: Parliament, Big Ben, Millenium
Eye, Millenium Bridge, Tower of London, or even just a bird,
interesting wall, pub, and more.
So it will be interesting to see if a photograph of a bird on a
railing of the river would get tagged with "Globe Theater" :)
Andrew
_______________________________________________
Geowanking mailing list
[email protected]
http://lists.burri.to/mailman/listinfo/geowanking
