Future of Search

I've always been of the mind-frame that the future of search/searching is in-fact incorporating image search. i.e the ability to submit an image and have the search engine report back with all the results it can find in the expanse (Intra/Internet). Video categorization and search would eventually be the next evolutionary step....but image search has to evolve and grow before that occurs.


I did post an idea back in 2006 in this very blog that did touch base upon the searching capability using and tagging entities and images. ( Click here and refer to item "xvi) Google Big-picture:")


Anyhow this is quick little post about a startup I discovered the other day. It's a company called IDEE (inc) which develops advanced image identification and visual search software. They are based out of Toronto and after reading about them initially, signing up for a beta account on their website and playing with the couple of products they have rolled out. After experiencing all of this, well let me just cut to the chase and state that I will be eagerly awaiting for their IPO to be rolled out.


Check out some of their products listed under their main page .


I spent some time checking out the capabilities of one of the application/service they offer. The service is called http://tineye.com. You have to register on the website to get on the beta (instantaneous). Basically it is an image recognition service whereby you upload an image on the website and hit go. In a typical submit/get request the idee servers will crawl through the image set and try and find an identical or near-identical match for the image you submitted. I ran some tests on the http://tineye.com website and here are some of the results I received for the various request (number of items returned):

1- tux (304 results submitted)
2- stanley cup (7 hits)
3- arctic fox (2 hits)
4- image of george bush (486 hits)
5- particular image of stephen hawking (14 hits)
6- particular image of Hal 9000 (2001 space odyssey) (37 results)


I'd be really curious to know how Idee's algorithm's actually work. The way I percieve it, it would obviously have to do with some sort of pattern recognition. But (and this is a presumption on my part) what would end up happening is each one of the images that Idee Inc's server's would crawl, they would carve out some sort of outline for these images and then store it in their back-end database. The outline would be the first step, the algorithm would also assign each one of the images some unique attributes that makes the image unique. Something along the lines of what facial features do for face-recognition. But this indexing and characterization is where the magic is occuring and that's what I would like to know is how it it occuring.

Otherwise if the algorithms are actually going down to the pixellization level then my theory goes to the bin. However...I highly doubt that the indexing is happening down on a pixellation level as the system/cpu overheading for rendering a job like this across the huge expanse of the internet would be too much for the servers to handle.


Already Idee Inc has some big names as their customers, including www.digg.com and Associated Press.


Do keep in mind that http://tineye.com is not the only service they offer :)


If image search is of interest to you, then you'd definitely want to track the progress of this company.


Before I go I have to mention that, I actually got to know about these folks reading the Toronto Star. Ironically now that I do a search on the toronto star website (www.thestar.com) there are no search results returned for either Idee and or tineye or the founding members of this company. Wasted 10 minutes trying to find the article I originally read in print format.