A familiar face, such as the face of a colleague or relative, is easily recognized despite changes in expression, viewing angle, and age. Face recognition processes are robust in the sense that a face memory can be triggered by a broad spectrum of perceptual stimuli. The many-to-one mapping of face stimuli to a single face memory may be described as a face representation's `attractor field.` According to this approach, a given face memory will be activated by any perceptual stimuli falling within the boundary of its attractor field. In this project the attractor field concept is investigated by probing the boundaries that separate competing face memories in recognition. Boundary testing is performed using an image technique known as morphing in which two face images are spatially averaged to produce a morphed face image. An attractor field boundary is defined as the point at which identification of the morphed image switches from one competing face memory to the other. The goal of the research is to understand how attractor fields might function in face and object recognition processes. This problem is approached first by determining the factors that influence the size of face memory's attractor field, and second by investigating how learning and experience might modify the properties of an attractor field. While the experiments are aimed at providing an account of attractor fields in face recognition, the proposal also addresses the role that attractor fields might play in general object recognition processes by examining their function in the recognition of non-face objects (i.e., cars, birds, dot patterns). Finally, the proposal investigates the computational principles of attractor field recognition in a series of neural network simulations.
