May 3, 2007: Face Recognition by Information Maximization
Filed in: IBM Lecture
Series
Dr.
Marni Stewart Bartlett, University of
California at San Diego
This talk will explore principles of unsupervised learning and how they relate to face recognition from the perspective of both machine vision and human visual perception. Dependency coding and information maximization appear to be central principles in neural coding early in the visual system. These principles may be relevant to how we think about higher visual processes such as face recognition as well. The talk will first review examples of dependency learning in biological vision, along with principles of optimal information transfer and information maximization. Next, the talk will review algorithms for face recognition by computer from a perspective of information maximization. The eigenface approach can be considered an unsupervised system maximizes information transfer only in the case where the input distributions are Gaussian. Independent component analysis (ICA) maximizes information transfer for a more general set of input distributions. Face representations based on ICA gave better recognition performance than eigenfaces, supporting the theory that information maximization is a good strategy for high level visual functions such as face recognition. Finally, the talk will review perceptual studies suggesting that dependency learning is relevant to human face perception as well, and present an information maximization account of perceptual effects such as the other-race effect and and face adaptation aftereffects.
Time permitting, at the end of the talk I will overview recent work at my lab on automatic facial expression analysis, and show a live demo.
Dr. Bartlett is Associate Research Professor at the Institute for Neural Computation, UCSD, where she co-directs the Machine Perception Lab. She studies learning in vision, with application to face recognition and expression analysis. She has authored over 30 articles in scientific journals and refereed conference proceedings, as well as a book, Face Image Analysis by Unsupervised Learning, published by Kluwer in 2001. Dr. Bartlett is Associate Editor for Neurocomputing, and she is also actively involved in the NSF Igert program on learning and vision in humans and machines at UCSD, as well as the NSF Science of Learning Center at UCSD on the temporal dynamics of learning. Dr. Bartlett obtained her Bachelor's degree in Mathematics in 1988 from Middlebury College, and her Ph.D. in Cognitive Science and Psychology from University of California, San Diego, in 1998. Her thesis work was conducted with Terry Sejnowski at the Salk Institute on face recognition by independent component analysis. She has also conducted research on automatic recognition of facial expression with Paul Ekman, and perceptual plasticity with V.S. Ramachandran. Her work spans machine vision and cognitive neuroscience, including papers in computer vision, visual psychophysics, neuropsychology, cognitive models of face perception, and the visuo-spatial properties of faces and American Sign Language.
Abstract
This talk will explore principles of unsupervised learning and how they relate to face recognition from the perspective of both machine vision and human visual perception. Dependency coding and information maximization appear to be central principles in neural coding early in the visual system. These principles may be relevant to how we think about higher visual processes such as face recognition as well. The talk will first review examples of dependency learning in biological vision, along with principles of optimal information transfer and information maximization. Next, the talk will review algorithms for face recognition by computer from a perspective of information maximization. The eigenface approach can be considered an unsupervised system maximizes information transfer only in the case where the input distributions are Gaussian. Independent component analysis (ICA) maximizes information transfer for a more general set of input distributions. Face representations based on ICA gave better recognition performance than eigenfaces, supporting the theory that information maximization is a good strategy for high level visual functions such as face recognition. Finally, the talk will review perceptual studies suggesting that dependency learning is relevant to human face perception as well, and present an information maximization account of perceptual effects such as the other-race effect and and face adaptation aftereffects.
Time permitting, at the end of the talk I will overview recent work at my lab on automatic facial expression analysis, and show a live demo.
Bio
Dr. Bartlett is Associate Research Professor at the Institute for Neural Computation, UCSD, where she co-directs the Machine Perception Lab. She studies learning in vision, with application to face recognition and expression analysis. She has authored over 30 articles in scientific journals and refereed conference proceedings, as well as a book, Face Image Analysis by Unsupervised Learning, published by Kluwer in 2001. Dr. Bartlett is Associate Editor for Neurocomputing, and she is also actively involved in the NSF Igert program on learning and vision in humans and machines at UCSD, as well as the NSF Science of Learning Center at UCSD on the temporal dynamics of learning. Dr. Bartlett obtained her Bachelor's degree in Mathematics in 1988 from Middlebury College, and her Ph.D. in Cognitive Science and Psychology from University of California, San Diego, in 1998. Her thesis work was conducted with Terry Sejnowski at the Salk Institute on face recognition by independent component analysis. She has also conducted research on automatic recognition of facial expression with Paul Ekman, and perceptual plasticity with V.S. Ramachandran. Her work spans machine vision and cognitive neuroscience, including papers in computer vision, visual psychophysics, neuropsychology, cognitive models of face perception, and the visuo-spatial properties of faces and American Sign Language.