Abstract: A feature-based elastic graph matching algorithm (FEGM) on the basis of the Lades' elastic graph matching face recognition algorithm is proposed in this paper. It includes two steps. First, the face feature such as eye, nose, mouth are localized by Gabor transforms, and a group of feature vectors composed by Gabor coefficients is extracted. Then the FEGM algorithm is performed to make the total cost function as small as possible, recognition result on Yale face database shows that recognition rate has been improved substantially.

Key words: face recognition, elastic graph matching, feature extraction, Gabor transform