When it comes to searching information on the web, there are plenty of tools that do a good job with text. When it comes to visual materials, the task is much harder.

Now researchers at the University of Pennsylvania have a new piece of software they think can learn to accurately tag pictures.

Recently, they tested ALIPR [Automatic Linguistic Indexing of Pictures] on 5,411 previously unseen images available on the popular picture-sharing site Flickr. For 51 percent of these images, the first word generated by ALIPR appeared in users’ tags. The program also produced at least one accurate word 98 percent of the time. The researchers employed images made publicly accessible by Flickr users, which were also openly accessible through Flickr’s own Application Programming Interface.

Anyone can give the system a try by going to the ALIPR site and entering the URL of a picture online or uploading one from a local computer.

The program then offers fifteen tags it “thinks” might be associated with the image. You can then help the program learn by selecting which of them might actually apply or adding your own.

In about a dozen tries with pictures of varying complexity, I found the software would usually come up with at least one tag that I might have chosen.

However, twice it matched on none and would often come up with some very strange choices, even in what I thought was a pretty obvious image.

Still, it’s a fascinating experiment in a field that has a lot of potential.

images, tagging, alipr