Eigenfaces for Recognition¶
Why this mattered¶
Turk and Pentland’s paper mattered because it reframed face recognition from a problem of hand-designed facial measurements or explicit 3-D reconstruction into one of statistical representation. By treating aligned face images as points in a high-dimensional image space and using principal components to find the dominant axes of variation, the paper showed that identity could be represented compactly as coordinates in a learned “face space.” This was a paradigm shift: recognition no longer required specifying eyes, noses, mouths, or geometric rules in advance. The system learned the useful basis directly from examples.
That made near-real-time face recognition practically plausible on the hardware of the early 1990s. The eigenface representation reduced comparison to a small vector of weights, enabling efficient recognition, incremental enrollment of new individuals, and a clear distance-based decision procedure. Just as important, the method linked computer vision, information theory, and cognitive neuroscience in a way that made face recognition look like a general pattern-recognition problem rather than a special-purpose biometric engineering task.
Its influence extended well beyond the specific PCA technique. Eigenfaces became a canonical demonstration that high-dimensional visual data could be made tractable by learning low-dimensional structure, helping set the stage for later appearance-based vision methods, subspace methods, Fisherfaces, manifold learning, and eventually data-driven deep face recognition. Modern systems no longer rely on linear eigenface spaces, but they inherit the central insight: effective recognition comes from learning representations from examples and comparing faces in that learned feature space.
Abstract¶
We have developed a near-real-time computer system that can locate and track a subject's head, and then recognize the person by comparing characteristics of the face to those of known individuals. The computational approach taken in this system is motivated by both physiology and information theory, as well as by the practical requirements of near-real-time performance and accuracy. Our approach treats the face recognition problem as an intrinsically two-dimensional (2-D) recognition problem rather than requiring recovery of three-dimensional geometry, taking advantage of the fact that faces are normally upright and thus may be described by a small set of 2-D characteristic views. The system functions by projecting face images onto a feature space that spans the significant variations among known face images. The significant features are known as "eigenfaces," because they are the eigenvectors (principal components) of the set of faces; they do not necessarily correspond to features such as eyes, ears, and noses. The projection operation characterizes an individual face by a weighted sum of the eigenface features, and so to recognize a particular face it is necessary only to compare these weights to those of known individuals. Some particular advantages of our approach are that it provides for the ability to learn and later recognize new faces in an unsupervised manner, and that it is easy to implement using a neural network architecture.
Related¶
- enables → A Global Geometric Framework for Nonlinear Dimensionality Reduction — Eigenfaces showed linear PCA embeddings for face images, setting up the contrast with Isomap's nonlinear manifold embedding.
- enables → Learning the parts of objects by non-negative matrix factorization — Eigenfaces showed that matrix factorization could learn compact face representations, while NMF replaced holistic PCA components with additive parts-based bases.
- cite ← A Global Geometric Framework for Nonlinear Dimensionality Reduction — Isomap relates to Eigenfaces through the shared use of eigenvector-based low-dimensional embeddings, but replaces linear PCA with geodesic manifold distances.
- cite ← Learning the parts of objects by non-negative matrix factorization — Lee and Seung contrast NMF's parts-based face representations with Eigenfaces' holistic PCA-based face representation.