Face recognition using patch manifold learning across plastic surgery from a single training exemplar per enrolled person

Although numerous methods have been developed for human face recognition, little investigation is focused on the human face recognition across plastic surgery and also single-exemplar face recognition. In this article, we present a new face recognition algorithm using patch manifold learning under plastic surgery conditions when only a single training exemplar per enrolled person exists. In the presented method, a face image is divided into a collection of patches which have no overlapping that are considered as a manifold. Then, we formulate face recognition under plastic surgery conditions using a single exemplar of each person as a problem of manifold-manifold matching to maximize the margin of manifold patches. A complete experimental investigation is done using the database of plastic surgery, AR and also FERET face databases. Experimental results indicate the superiority of the presented algorithm for face recognition in single-sample databases.