Something that comes naturally to humans is still a challenge for computers: recognizing images. We’re able to learn what something or someone looks like. We can look at an image or sculpture of a subject and recognize what it is. Even if the image isn’t perfect, we can figure it out. Computers aren’t as good at that. That’s part of the reason classic CAPTCHA tests work — they’re easy for us to recognize but computers find it tricky.

But computers are getting smarter. Even the inventors of CAPTCHA tests don’t see this as a bad thing — when a computer beats a CAPTCHA test, it means we’re one step closer to artificial intelligence. A failure of one system is a triumph for another.

Several teams around the world are working on improving computers’ abilities to identify images. With reliable image recognition software, we’d have an  improved search capability (most image searches rely on meta data or text captions, not the actual images). You could catalog billions of images automatically with a strong enough system.

So how do we do that? How do you teach a computer how to see? At MIT, scientists have worked on this problem for some time. Their approach is to teach the computer to look for patterns and sequences in numerical data. Each image is composed of pixels. The computer interprets the pixels as a series of numbers. If the computer can identify similar numerical series in multiple images, it can recognize that the images are all of the same subject.

Beyond cataloging information, there are other potential applications for image recognition software. European researchers developed software that could recognize images of buildings. Take a photo of a building in a researched city and the software will retrieve information about it. It’s combining digital photography with augmented reality. The software doesn’t work by magic though  — it uses a huge database.

Image recognition software could help revolutionize cryptography. One form of cryptography is steganography — hiding messages. Steganography can take many forms. One of those is to hide a message within an image. A computer that can analyze and recognize images might be able to discover a hidden message faster than a human. Of course, you could always hide the message within the code of the image itself — image recognition software would be useless in that case.

Image recognition software will play an important role in robotics as well. As we develop more sophisticated algorithms, robots will be able to analyze and recognize elements in their environments. Which means we’re one step closer to me getting that robot butler I’ve always wanted.

Learn more at HowStuffWorks.com:

How Robots Work
How CAPTCHA Works
How Code Breakers Work