May 2005
S M T W T F S
1 2 3 4 5 6 7
8 9 10 11 12 13 14
15 16 17 18 19 20 21
22 23 24 25 26 27 28
29 30 31        

« Runaway Bride Jennifer Wilbanks Becomes Blog Martyr | main | Google Web Accelerator: Completing the Trifecta of Web Domination »

Content-Based Image Retrieval Finds De Plane! De Plane!

tattoo.jpg

What if you could search for images not by keywords, but by using the images themselves? Content-Based Image Retrieval (CBIR) is the process of applying a 'computer eye' to the image retrieval process, making searching for images in large databases easier. The search looks at the contents of the images themselves rather than human-controlled metadata such as captions and keywords.

In plainer terms, this means instead of entering text to find images, you click on an existing image in order to view related ones. If properly executed, the process is much more intuitive (how do you classify Rover - is he a dog or an animal?), scalable (how do you assign values to hundreds of hours of surveillance photos?), and customizable (you can create specialized CBIR for each query or domain) than the way we search for images today.

Right now there's a CBIR research project going on at Penn State that's using the largest aviation photo database (think tons of pictures of planes) in the world as its guinea pig. Some queries are way off, but it's not hard to see where this may eventually be going. If Yahoo! or Google got their hands on this, they'd have an immediate leg up on other image searches.

The user discussion on Slashdot raises a few good points.

  • It's still quite crude - one user looking for an image of the Qantas logo said, "It seems like it's simply looking for similar patches of color...Interesting, but not ready for public consumption just yet."
  • Let's get sue happy - this could become a tool for companies looking to police their intellectual property, with firms like Disney using it to do a sweep for all instances of Snow White, whether it's a direct copy or not.
  • It's been done before - as one Slashdotter points out, the first version of Google Image Search had a "find similar" link next to every image. After a few months the feature was disabled due to increasingly inaccurate results.

Posted by RM at May 5, 2005 10:43 AM

Send to a friend


Tagging

| |


Trackback Pings

TrackBack URL for this entry:
http://searchviews.com/mt/mt-tb.cgi/1783


Comments

Post a comment




Remember Me?