Brain and Cognitive Sciences

My general area of research is the study of vision - including the processing of visual information by the human visual system, and computer vision. The goals of this research are to understand how our own visual system operates, and how to construct artificial systems with visual capabilities including, for example, aids for the visually impaired. The two goals, understanding human vision and computer vision, are strongly interconnected. At present, the performance of the human visual system is superior in almost every respect to that of machine vision systems. This is particularly striking in the case of object classification, where the performance of the best computer models cannot rival the performance of a three year old child. The study of the computations performed by the human visual system can therefore lead to new insights and to the development of new and better methods for analyzing visual information. At the same time, given the enormous complexity of the human visual system, a better theoretical understanding of the computations underlying the processing of visual information can supply useful guidelines for empirical studies of the biological mechanisms subserving visual perception. The focus of my current research in on the topic of visual object recognition, and the biological modeling of information processing in the visual cortex. An example of recent research direction is work on a fragment-based approach to object classification and segmentation. In this approach objects within a classe represented in terms of common image fragments, that are used as building blocks for representing a large variety of different objects that belong to a common class, such as a face or a car. Optimal fragments are selected from a training set of images based on a criterion of maximizing the mutual information of the fragments and the class they represent. These fragments are typically of intermediate complexity in size and resolution. This approach appears to have good classification and generalization capabilities. Image segmentation is obtained in this approach as a part of the segmentation process, and it combines both bottom-up and top- down components.