Skip to main content

Google develops computer vision accurate enough to solve its own CAPTCHAs

Google develops computer vision accurate enough to solve its own CAPTCHAs

Share this story

Google has cracked the CAPTCHA. In a paper published this week, Google researchers say that they've developed an algorithm that can accurately solve Google's own CAPTCHA puzzles — those obfuscated jumbles of letters and numbers you type in on websites to prove that you're human — with 99.8 percent accuracy, obviously posing something of a problem to the puzzle's intended purpose of weeding out robots. The new system was developed to help Google automatically analyze hard-to-read signs and house numbers photographed by its Street View cameras, allowing it to accurately match images with locations on a map.

"Relying on distorted text alone isn’t enough."

Despite being near perfect when it comes to CAPTCHAs — a feat that plenty of humans can't even manage — the new system's analysis of Street View imagery isn't quite as accurate, correctly identifying the text just over 90 percent of the time. When analyzing house numbers specifically, however, its accuracy jumps up to over 96 percent.

These CAPTCHAs have been correctly solved by the new algorithm.

Google is, of course, specially suited to developing such advanced automated text analysis because of its extensive work with Street View and reCAPTCHA, its own CAPTCHA service. Even so, Google says that it's already found ways to further protect reCAPTCHA from being broken by others' computers. "Thanks to this research, we know that relying on distorted text alone isn’t enough," Vinay Shet, reCAPTCHA's product manager writes in a blog post. Shet explains that part of this is analyzing a user's full interaction with the CAPTCHA puzzle — and not solely whether they can get the answer right.