Abstract:

Abstract: Nowdays computers solve so many difficult tasks for us that we take them for granted. They can automatically land an airplane, detect a problem in your car, create special effects for movies, write music, and understand your questions over the telephone. Ask yourself, why is it that computers can predict global weather patterns yet, if you plug in a camera, they cannot tell a dog from a cat? The list of seemingly trivial visual tasks that computers cannot do goes on and on. For example, humans easily recognize their friends at a distance of 5-10 meters. Most people have no problems locating objects around them. We effortlessly percieve 3D shapes from just looking at photos or paintings. We can do all of this because our brain analyzes images projected onto our eyes' retinas in real-time. "Computer vision" is an area of computer science devoted to making computers understand images, much as humans do. Images come from a digital photo/video cameras connected to a desktop PC, laptop, or cellphone. They also come from MRI/CT scanners, electronic microscopes, ultrasound sensors, and many other sources. This talk will present some of the challenges facing computer vision and some of its recent advances. We will also discuss some mathematical models widely used in computer vision. In particular, discrete models based on Markov Random Fields and related models from based on deferential and integral geometry. We will also discuss some related optimization issues.