Research Merit: the research proposal is to develop a model of region segmentation based on local motion signals, building on a model by Yuille and Grzywacz. The basic idea is that local motion signals form the basis of a high dimensional representation of the motion, and the data in this space can be grouped into regions on the basis of their propinquity in this space. The model has some interesting features, one of which is the notion of accrual of confidence over the course of an object's trajectory. This information is used in making the region decisions. Motion templates are also used, and this is a nice way to incorporate the observer's past experience. The system minimizes the number of regions, maximizes their smoothness, and maximizes the fit with the motion templates. This is a very interesting and plausible model. The candidate will implement the model and develop estimates for some critical parameters. Simulations will be performed to fit a variety of critical results in the literature. In addition, some new psychophysical experiments will be performed to investigate human observers' ability ot segment on the basis of motion statistics, and whether it is possible to recursively form higher-order regions. These experiments are interesting, but the primary strength of the proposal is in the development of the model.
