Picli is the latest addition to the web’s social recommendation networks. Usual stuff: submit, vote up or down, become popular or not.

Nothing new, except for the fact that Picli is all about images.

I’ve been browsing around there a little bit, obviously with SEO in mind (or in our case Canadian SEO). A nice place which already gets some good activity and which clearly presents multiple ways in which it can be used to promote a web site.

At the University of California, San Diego (UCSD), they have been working on a way to label, to identify, items and subjects in photos. The system is called Supervised Multiclass Labeling (SML). Supervised because human action is involved. Multiclass because the system is able to identify more than “label” in any image.

To teach the system to recognize a mountain it is shown a multitude of photos with a mountain. Can be any mountain from any angle with any feature. The photos don’t need to be similar in any way except for that each does have a mountain in it.

During this process the computer divides each photo into 8-by-8 pixels squares and notes information for each square. Collectively these are called a “bag of features” and from that bag the researchers choose which information is most consistent, most overlapping.

After that you have to test the system. That is best done on a large image database and here is where it gets interesting from an SEO point of view.

Dr. Pedro Moreno is one of the authors of the UCSD SML paper. Dr. Pedro Moreno also works at Google.

Started in 2004 SML was tested by the authors at Google this Summer, running on a cluster of 3000 Linux machines.

Google also is the owner of Neven Vision, a company which holds the patents to facial recognition in both static images and videos.

Who knows what we’re helping to verify by playing Google Image Labeler.

I hang out at Twitter where I enjoy the company, the buzz, the nuggets of info and opinion we pass along.
Join me on Twitter!

Leave a Reply