DailyDirt: Will Computers Have 20/20 Vision?

Algorithms for image recognition are getting better all the time, but computer vision is still very different from how humans look at images. Computers aren't capable of describing an image as well as a typical 5-year-old, but they can sift through millions of images before a kid can blink. Here are just a few examples of algorithms getting better at seeing the same things that we see.

• Computers could play "Where's Waldo?" using a variety of techniques. Is there an optimal way to play Where's Waldo? Perhaps, but where's the fun if you use an algorithm? [url]
• The Viola-Jones algorithm is commonly used in cameras for face detection, but it's a relatively simple software trick that only picks up full faces that are looking directly at the camera. More advanced algorithms can now spot a face at an angle and even when the face is partially blocked. An approach called the Deep Dense Face Detector can even recognize upside-down faces, using a neural network technique, and it might vastly improve automated image recognition. [url]
• A visual Turing(-like) test would challenge algorithms to evaluate images compared to human visual recognition. As with the original Turing test, the passing grade for artificial intelligence would continually change as AI gets better. [url]

If you'd like to read more awesome and interesting stuff, check out this unrelated (but not entirely random!) Techdirt post via StumbleUpon.

Permalink | Comments | Email This Story