Re: Searching with Images instead of Words

2005-01-16 Thread Bill Stewart

Expecting a front view of an image to match with a
side view of the same image is impossible. They are
both disjoint sets of information.
If all the images are frontal images, we can match
them with a hight probability, otherwise I doubt this
technology has a future.
I think it definitely has a future.
I'm a bit skeptical about whether it's a _near_ future, though
It sounds especially possible for specific classes of pictures,
such as outdoor locations in major cities.



Bill Stewart  [EMAIL PROTECTED]  



Re: Searching with Images instead of Words

2005-01-14 Thread Sarad AV
hi,

They had been researching on this line in Indian
Institue of Science, Bangalore. I think image
searching has fundamental limits. For  successfully 
matching two images, there should be a subset of
information in both that totally match or match with a
high probability.

Expecting a front view of an image to match with a
side view of the same image is impossible. They are
both disjoint sets of information.

If all the images are frontal images, we can match
them with a hight probability, otherwise I doubt this
technology has a future.

Sarad.


--- Eugen Leitl [EMAIL PROTECTED] wrote:

 
 Link:
 http://slashdot.org/article.pl?sid=05/01/13/184226
 Posted by: CmdrTaco, on 2005-01-13 20:29:00
 
from the blessing-for-those-who-can't-spell dept.
[1]johnsee writes A computer vision researcher
 by the name of Hartmut
Neven is [2]developing ingenious new technology
 that allows the
searching of a database by submitting an image,
 for example, off a
mobile phone camera. Imagine taking a photo of a
 street corner to find
out where you are, or the photo of a city
 building to see its history
 
IFRAME: [3]pos6
 
 References
 
1. http://www.sandstorming.com/
2.

http://www.thefeature.com/article?articleid=101341ref=5147543
3.

http://ads.osdn.com/?ad_id=2936alloc_id=13732site_id=1request_id=9329739
 
 - End forwarded message -
 -- 
 Eugen* Leitl a href=http://leitl.org;leitl/a

__
 ICBM: 48.07078, 11.61144   
 http://www.leitl.org
 8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443
 8B29 F6BE
 http://moleculardevices.org
 http://nanomachines.net
 

 ATTACHMENT part 2 application/pgp-signature 





__ 
Do you Yahoo!? 
The all-new My Yahoo! - Get yours free! 
http://my.yahoo.com 
 



Re: Searching with Images instead of Words

2005-01-14 Thread Tyler Durden
Expecting a front view of an image to match with a
side view of the same image is impossible. They are
both disjoint sets of information.
If all the images are frontal images, we can match
them with a hight probability, otherwise I doubt this
technology has a future.
You are applying pure logic to a very complex subject. I'd bet this is 
already routinely done by TLAs and whatnot, at least as a pre-screen before 
human photograph inspectors.

The most obvious hole in your statement is with respect to 2D Spatial FFTs 
of the image...you can probably greatly increase your match probability via 
certain masking criteria applied to the 2D FFT. And from there there's lots 
of stuff that can be done with colors and other indirect stuff such as 
(perhaps) camera signatures in the photo (eg, If there's text that says 
Hamamatsu Synchroscan Streak Camera then don't bother doing the FFT--it 
ain't a picture of your dog).

Look...a human being can recognize the side image of a person a lot of the 
time. There should be no reason this intelligence can't be encoded somehow.

-TD