Professor Donald Geman and Xiaodong Fan, a doctoral student, are working on a project in automated visual recognition. The general problem is to detect all instances from a repertoire of object classes in complex, highly cluttered, gray-level scenes. In this particular project, the objective is to identify the characters on a license plate based on data from a photograph of the rear of the car. Two results are shown above. This is easy for human beings but remains very difficult for computer vision systems. The main focus is sequential learning. As in biological vision and learning, one wishes to build systems inductively from data and in a sequential manner, efficiently adapting existing mechanisms as new examples come along. In this work, the architecture that is updated is a tree-structured hierarchy of tests designed for rapidly detecting individual characters.