Object-recognition’s Extreme Close-up

Researchers at MIT and UCLA have developed an object-recognition system that may revolutionize the computer vision field. The system takes a “bottom-up” approach to recognizing objects, first identifying the smallest features it can. Then, it searches for places where those small features are repeated and builds increasingly sophisticated shapes from there. Over time, the system builds a “hierarchical catalogue” of the image’s parts—with the top layer being the entire analyzed object.

According to TechNewsDaily, the researchers believe this approach will make object-recognition systems easier to build and enable them to eat up less computer memory.