Sorry for the lull in posting but extensive deal-screening, M&A activity (to be announced), skiing in Zermatt and the usual Christmas gluttony have interfered with blog posting. However, today I have another $1 billion idea for you that is just waiting to be built. Look closely at these two images: Google products by unique monthly visits, 2007 vs 2006 and associated growth rates.
Try and finding an image of Zermatt on Google Image. It’s an excruciatingly painful experience.
Image search is extremely popular. And yet it is really, really bad. The reason is of course that the query is text/semantic and the raw indexed object is binary. So you have Google, Yahoo and new players like Picsearch and Pixsy turning the metadata around the image into an index for those images and you can see very well on Google Image Search that the alt/title tags and captions have a tenuous relationship at best with the actual content. This means the search results, well, suck.
The ginormous problem in image search is the correct labelling of unannotated images. This is even more difficult than video since there is no speech-to-text (e.g., Blinkx) or multi-angle views (e.g., Viewdle). If you can solve automagic image annotation - and believe me, many many university researchers are working on the problem and are offering their tech for commercialization - then you are sitting on a multiple billion dollar startup.
So, how would you start going about doing this? I would research who leads in computer vision across Europe. I would look at news such as these and read (and understand) the papers and get in touch with the relevant authors. And I would start designing a beautiful interface to pitch to the lab coats as a front-end for their technology. Even computer science folks get excited by rounded edges.
In the end, who could resist working for and who could decline financing a promising company that has as its mission “Organizing the World’s Images”? It’s big, audacious, hairy and gutsy. And you could really change the world if you succeed. Good luck!
This work is licensed under a
Creative Commons Attribution-Share Alike 2.0 License. Please attribute. And do contact me if you are looking for financing at max DOT niederhofer AT gmail DOT com.
3 Comments
A friend of mine works at polarrose now. Seems to be what you are suggesting but with a focus on people :) They seem to be well financed though.
There are MANY startups working on this problem: http://www.inperc.com/wiki/index.php?title=Image_search#Visual_image_search_engines. Very little success…
@Oliver: of course I know Polar Rose and Nikolaj quite well. I like it a lot but what I want is automatic annotation of everything, not just faces. Also, I would love to be able to use something like Reuters or Getty as a first start rather than rely on manual tagging.
@Peter: Yes, I know most of those. The fact that many folks are working on it is definitely a sign that there is a market need for this stuff. However, none of those listed really do what I want: automatic labelling of unannotated images.