Visual search lets you find what you’re looking for, literally

A new search engine,, “finds things that look similar”. That is, it lets users search for images that resemble a target image. At least, so long as the images in questions are ones of upmarket consumer goods. The categories on the site are relentlessly Sex-in-the-City commercial, such as finding a watch that looks like the one worn by Paris Hilton at the “Street Sexy” fashion show. Click here to try that particular search – gotta say the initial results look pretty random to me, with big and small watches, some with chronometer-type inset dials, some without, some with metal bracelets, others with leather, etc. Then what do I know about fashion? But there are some interesting features to fine-tune the initial search results, such as prioritising a particular part of the image, or prioritising general factors such as colour, shape or pattern.

Picture 2

Unfortunately, doesn’t let you point to an arbitrary image and then find similar images out there on the web. Instead it just lets you browse through predetermined collections of bling.

Apparently when it was trialled on MySpace, this kind of “visual similarity shopping” was much more popular than facial recognition search, which surprised me at first. I’d have thought there was quite a market for finding images that looked like Paris Hilton (e.g. with user controlled parameters for degree of unclothedness), but perhaps that market is already well satisfied on the web.

There must, however, be interesting applications of this underlying technology beyond yet-more-shopping: what about a medical diagnosis engine? Or at least a search engine to find other similar medical or other technical images? Surely it would be really useful for researchers to be able to compare micrographs or any other kind of specialist images with either the content of the web or a specified database.

Technorati Tags:



LinkedIn button
free debate

RSS feed for this blog

Subscribe via email



Top Clicks

  • None

%d bloggers like this: