Building Computers with Vision

As smart as computers are these days, they still have crummy eyesight. Electronic eyes cannot readily distinguish different shapes, let alone add context to what they see. Humans, on the other hand, take it for granted that we can not only instantly tell a cat apart from a dog, but can identify faces of individuals on sight.

Now, a team at MIT's Center for Biological and Computational Learning is seeking to develop software that could help electronic cameras understand more of what they see, and act on it instantly. For instance, such software could be added to surveillance cameras to allow them to instantly identify an intruder and sound an alarm. Or, it could help a robot more clearly navigate visually.

Of course, this is no mean feat, as humans use as much as 40% of their brain function for interpreting what they see. Moreover, we don't fully understand how the brain works in this area. The software developers are seeking to replicate this action by helping computers analyze images pixel by pixel, and identifying differences among pixel positions and colors.

Source: MIT Technology Review