This paper presents a system for recognizing human faces from single images in a database containing one image per person. The system addresses the challenge of image variation due to differences in facial expression, head pose, position, and size. It uses a novel approach called Elastic Bunch Graph Matching (EBGM) to extract concise face descriptions in the form of image graphs. These graphs are constructed from a small set of sample image graphs and are used to represent faces in a way that is invariant to variations in size, position, and orientation. The system uses Gabor wavelets to extract local features, which are robust to lighting changes and small shifts. Jets, which are sets of wavelet coefficients, are used to describe fiducial points on the face. The system then matches these jets to new images to generate image graphs, which are then used for recognition. The system is tested on the FERET and Bochum databases, including recognition across different poses. The system's core is a labeled graph, where nodes are labeled with wavelet responses and edges are labeled with distance information. Model graphs are stored and matched to new images to generate image graphs. These image graphs are then compared with model graphs to find the best match. The system is robust to small in-depth rotations of the head and can handle larger rotations with different graph structures and designer-provided correspondences. The system uses a combination of jets and distances to represent faces, allowing for accurate recognition even when faces vary in expression, pose, and size. The system is flexible and can be applied to other in-class recognition tasks, such as recognizing individuals of a given animal species. It does not require extensive training for new faces or object classes, only a moderate number of examples to build a bunch graph. The system is efficient, with recognition taking only a few seconds per image. It is robust to variations in facial expression and can handle rotations up to about 22 degrees. However, it is less effective for larger rotations.This paper presents a system for recognizing human faces from single images in a database containing one image per person. The system addresses the challenge of image variation due to differences in facial expression, head pose, position, and size. It uses a novel approach called Elastic Bunch Graph Matching (EBGM) to extract concise face descriptions in the form of image graphs. These graphs are constructed from a small set of sample image graphs and are used to represent faces in a way that is invariant to variations in size, position, and orientation. The system uses Gabor wavelets to extract local features, which are robust to lighting changes and small shifts. Jets, which are sets of wavelet coefficients, are used to describe fiducial points on the face. The system then matches these jets to new images to generate image graphs, which are then used for recognition. The system is tested on the FERET and Bochum databases, including recognition across different poses. The system's core is a labeled graph, where nodes are labeled with wavelet responses and edges are labeled with distance information. Model graphs are stored and matched to new images to generate image graphs. These image graphs are then compared with model graphs to find the best match. The system is robust to small in-depth rotations of the head and can handle larger rotations with different graph structures and designer-provided correspondences. The system uses a combination of jets and distances to represent faces, allowing for accurate recognition even when faces vary in expression, pose, and size. The system is flexible and can be applied to other in-class recognition tasks, such as recognizing individuals of a given animal species. It does not require extensive training for new faces or object classes, only a moderate number of examples to build a bunch graph. The system is efficient, with recognition taking only a few seconds per image. It is robust to variations in facial expression and can handle rotations up to about 22 degrees. However, it is less effective for larger rotations.