When I researched an article on multimedia search last year for EContent Magazine, (The resulting article was republished on Streamingmedia.com last December.) I learned in the course of my research that it's hard to search for non-text elements because they lack the contextual language of text. Seems logical enough, but the way most search engines get around this is by using the text-based metadata around the image or video to get searchers in the right neighborhood. It works in a 1990s sort of way, but what the world really needs is more advanced multimedia search.

That's why my eyes popped a bit when I came across this NYT article this morning while scanning today's technology news. It seems Google is experimenting with image recognition to provide a more advanced way to search for images (and one assumes eventually videos). The problem is that this is so resource-intensive, according to the article, that Google can only work with a small sub-set of its huge image repository. And if it's too resource-intensive for Google, you know we are talking about some serious resources.

Google is hoping to do for images, what page rank once did for text with its original search algorithm that rocked the world all those years ago. We shall see where this goes, but for now, it's interesting to see that Google is at least playing around with this, and as processor power and computer knowledge increases, we should begin to see major break-throughs around this type of search technology. For now, we are stuck mostly with metadata and some other interesting approaches outlined in the Streamingmedia.com article, but this announcement certainly bodes well for the future of mutimedia search.

This may be one area where Google is beat. A little program called OutWit Images has pioneered this field and has a decent following of its clever Catch function.

This may be one area where Google is beat. A little program called OutWit Images has pioneered this field and has a decent following of its clever Catch function.

Google should seriously think of buying tineye to enhance its services

This is an interesting concept and Google is correct about the amount of resources it will need. Its not just about resources its about resources and speed and very good initial algorithms. The complexity is huge, this is the main problem in a nut shell. Think about it in terms of each picture analysed google need to analyse one picture in about 1000 of a second but maybe even 1000,0000 of a second if not and lets make this simple say they analyse on 1 picture per second then in 60 minutes they have only analysed 60 pictures. Therefore even if you are using a supercomputer or several suppercomputers it would take google for ever to analyse all the trillions of photos it has crawled over the years. Remember its a search engine so in theory if it stored the pictures on its own servers I guess it could be breaching copyright. Therefore as we know pictures move around get deleted and are added to peoples web-sites every day. Therefore it would have to constantly be able to re-tag trillions of photos all the time. Its initial search would have to be as good as the "General Picture Recognition Software" that has a very good initial find methodology if your interested in trying it out goto http://www.generalpicturerecognition.com/