Face recognition by elastic bunch graph matching

This paper presents a system for recognizing human faces from single images drawn from a large database that contains one image per person. Faces are represented by labeled graphs based on a Gabor wavelet transform, and image graphs of new faces are extracted through an elastic graph matching process, after which they can be compared using a simple similarity function.

The system improves on the earlier approach of Lades et al. (1993) in three respects: it uses phase information for accurate node positioning, it employs object-adapted graphs to handle large rotations in depth, and it bases image graph extraction on a novel data structure called the bunch graph, which is constructed from a small set of sample image graphs. These contributions enabled more robust matching of new faces against a large single-image face database.