Finding an edge
Determining the boundaries of objects is one of the central problems in computer vision. It's something humans do with ease: We glance out the window and immediately see cars as distinct from the sidewalk and street and the people walking by, or lampposts as distinct from the facades of the buildings behind them. But duplicating that facility in silicon has proven remarkably difficult. One of the best ways for a computer to determine boundaries is to make lots of guesses and compare them; the boundaries that most of the guesses agree on are likeliest to be accurately drawn. Until now, that process has been monstrously time consuming. But Jason Chang, a graduate student in the Department of Electrical Engineering and Computer Science, and John Fisher, a principal research scientist at MIT's Computer Science and Artificial Intelligence Lab (CSAIL), have figured out how to make it at least 50,000 times...