CVPR 2001 Short Course

 

 

Face Recognition by Humans & Machines: A Tutorial Survey

 

Instructor: Baback Moghaddam (Mitsubishi Electric Research Laboratory)

 

Duration: 4 hours

 

COURSE DESCRIPTION

 

The purpose of this short course is to give the uninitiated as well as those just starting in the field of face recognition, an historical overview of automatic computerized face recognition as well as the physiological and cognitive aspects of face recognition by primates and humans. This being CVPR, a greater emphasis will be placed on computational techniques, drawing parallels with neurophysiological findings, wherever appropriate. The ultimate goals of this short course are to understand what’s so special about faces, just how hard is face recognition by machines, and why is face recognition so important in the coming age of ubiquitous cameras and unprecedented surveillance.

 

Specifically, the course will:

 

* highlight the special role of this important visual routine in the context of sociological interaction and the ultimate survival of higher species dependent on recognizing their individual members.

 

* provide some insights based on recent findings as to how the primate brain has specialized for this visual task. In particular, we’ll examine the role of cortical areas IT and STS and the wealth of scientific knowledge gained by single-cell recordings and cortical ablation in primates as well as the study of performance deficits in humans where these "specialized areas" are damaged by stroke and head trauma.

 

* present a comprehensive survey of computer vision methods for face recognition dating back from the 1960s to the current. The various methods covered will include: feature-based, appearance-based, view-based, 2D shape-texture models, 3D shape-texture models, as well as various subspace representations (PCA, LDA, ICA, etc.) and different matching algorithms (Lp k-NN, Bayesian, etc.)

 

* present an overview of DARPA’s FERET program, its goals, testing methodologies and competition results and why it has been so critical to recent progress in the field

 

* discuss the shortcomings of current state-of-the-art techniques and discuss what can be done to overcome them. Some particular issues tackled are pose, illumination, non-rigid deformation (expression) and the pros and cons of 3D vs. view-based modeling.

 

 

BIOGRAPHY

 

Baback Moghaddam is a Research Scientist at Mitsubishi Electric Research Laboratory (MERL) working primarily in the areas of computational vision and image processing. His current research interests are probabilistic visual modeling for object recognition, facial analysis, statistical learning theory and advanced pattern recognition techniques for biometrics. Prior to MERL, Dr. Moghaddam was at the Vision & Modeling Group at the MIT Media Laboratory where he developed the MIT automatic face recognition system that won the 1996 DARPA "FERET" face recognition competition. Dr. Moghaddam is a member of the IEEE and ACM.

 

 

CONTACT

 

Baback Moghaddam

Research Scientist

Mitsubishi Electric Research Laboratory

201 Broadway 8th floor

Cambridge, MA 02139

Voice: 617-621-7524

Fax:   617-621-7550

 

 

Back to Main