Seeing Eye to Eye: Perspective Adjustment for Videoconferencing

The quality of communication in videoconferencing is compromised by the fact that participants cannot maintain eye contact. We propose an original solution to this problem. Using color and depth information from a Kinect camera, we will use a Conditional Random Field to model each participant’s eye position in 3D space, which will allow us to reorient the image to create the illusion of looking through a window. Finally, we will use a Fields of Experts model to fill in unobserved parts of the transformed image.