Google has released research results about a new test to foil computers pretending to be humans by requiring them to orient an image so it's upright.
A persistent problem on the Internet is screening out automated computer systems that can be used, for example, to sign up for spam-sending e-mail accounts or post comments designed to improve a site's search results. Google, which already devotes a lot of resources to block e-mail and Web spam, has tried a new test to keep the bots at bay.
The test is the latest variation on a screening technique called a Captcha (completely automated public Turing test to tell computers and humans apart). The idea is that people can often tell which way is up in a photo, but computers have a harder time.
Captchas are in widespread use today, usually in the form of obscured or distorted text that people can still read. But there's a lot of work in the area, including and distinguishing between cats and dogs.
Here's how Google authors Rich Gossweiler, Maryam Kamvar, and Shumeet Baluja described the image-orientation technique in their paper (click for PDF):
This task requires analysis of the often complex contents of an image, a task which humans usually perform well and machines generally do not.
Given a large repository of images, such as those from a web search result, we use a suite of automated orientation detectors to prune those images that can be automatically set upright easily. We then apply a social feedback mechanism to verify that the remaining images have a human-recognizable upright orientation.
The main advantages of our Captcha technique over the traditional text recognition techniques are that it is language-independent, does not require text-entry (e.g. for a mobile device), and employs another domain for Captcha generation beyond character obfuscation. This Captcha lends itself to rapid implementation and has an almost limitless supply of images.
We conducted extensive experiments to measure the viability of this technique...Our Captcha technique achieves high success rates for humans and low success rates for bots, does not require text entry, and is more enjoyable for the user than text-based Captcha.
The tricky part is finding the right balance between too easy and too confusing. Some images are hard for people to orient correctly, and some have cues--faces, text, blue skies, and green grass--that computers can use to figure out which way is up.
To get around this issue, while being able to draw from the large number of images on the Web, the technique presents people with new images as well as those known to perform well. If people have trouble consistently telling which way is up, that image isn't included in the library.
The researchers like their system in part because the image doesn't have to be obscured or distorted, as in text-based Captchas such as those Google currently employs. But image-based Captchas aren't immune from the bot vs. Web site arms race.
"As advances are made in orientation detection systems, these advances will be incorporated in our filters so that those images that can be automatically oriented are not presented to the user," the researchers said. "The use of distortions may eventually be required."