TR#245: View-Based and Modular Eigenspaces for Face Recognition

Alex Pentland, Baback Moghaddam, Thad Starner

Appeared in:
IEEE Conference on Computer Vision & Pattern Recognition,
Seattle, Washington, June 21-23, 1994.

In this work we describe experiments with eigenfaces for recognition, verification, and interactive search in a large-scale face database. Accurate visual recognition is demonstrated using a database of O(10^3) different faces. Questions of overall accuracy and sensitivity to orientation, scale, illumination are also addressed. In particular, the question of recognition under changes in viewing orientation is examined in detail. A view-based multiple-observer eigenspace technique is proposed for use in face recognition under variable head orientations. In addition, our recognition system is augmented by a modular eigenspace description which incorporates salient facial features such as the eyes, nose and mouth, in an eigenfeature layer. This modular representation yields a higher recognition rate as well as more robust framework for recognition. A robust automatic feature extraction technique using feature eigentemplates is also demonstrated.