Sachin Gangaputra, a doctoral student in CIS, is working with Professor Donald Geman to automatically locate human faces in still photographs as well as track them through video sequences. They use a hierarchy of statistical models to capture the variation in face appearances. There is a progression from models which accommodate very general poses but are not highly discriminating to those which are dedicated to speci.c poses but only rarely generate errors. Image data is then examined by traversing the hierarchy in a breadth-.rst, coarse-to-.ne manner. In this way, most image regions are dismissed as containing faces with very little computation. One extension to video sequences is based on a dynamic, tree-structured model which is Markov in both space and time. The still-frame algorithm is combined with constraints derived from temporal continuity and other cues, such as color, in order to track faces through multiple frames with very few missed detections and false positives.