Archive for November, 2009

Searching for "similar images"

Monday, November 2nd, 2009

We’ve discussed content-based image search before, but that was three years ago. Since then, the technology has matured into a useful everyday tool.

Suppose you have an image, and you want to find others like it. You can use the TinEye image search engine where you can either upload an image file, or submit a URL. TinEye will then display a list of similar images.

TinEye claims to index over a billion images, but it doesn’t always find a match. The matches that it does find tend to be very good. In particular, it will find images that include any one of the major components of the original image.

The other alternative is Google Images, which has rolled out a “Find similar images” option. Perform a regular image search, and you will find that most of the result images have a “Find similar images” link beneath them.

Google finds a different kind of match than TinEye. TinEye is quite literal, whereas the Google matches are more broad. Google seems to be taking account of the textual context on the image’s target page, as well as the visual characteristics of the image. TinEye seems to be matching some kind of literal measurement of the image components, such as their angle and height/width ratio.

Google’s interface is easier to use, because you can keep refining your search by clicking “Find similar images” on your best match so far. TinEye requires you to start each search afresh, although they also offer a browser plugin for easy searching on any image that you find on any webpage.

As at November 2009, TinEye requires free registration. Google Images is free to use without registration.

Similar images found by Google Image Search (left column) and TinEye (right column)

Similar images found by Google Image Search (left column) and TinEye (right column)