Mass-preserving maps for registration and visual tracking

We consider a new method for an important aspect of the visual tracking problem. Tracking in the presence of a disturbance is a classical control issue, but because of the highly uncertain nature of the disturbance, this type of problem is very difficult. A key issue in many visual tracking tasks is that of registration. Image registration is the process of establishing a common geometric reference frame among several data sets taken at different times. In this paper, we propose a method of registration based on the Monge-Kantorovich problem of optimal mass transport. We argue that such an approach can also be very useful for several problems in controlled active vision