Self-adaptive classifier fusion for expression-insensitive face recognition

We address a self-adaptive face recognition scheme which is insensitive to facial expression variations. The proposed method takes advantage of self-adaptive classifier fusion based on facial geometry and RBF warping technology. Most previous face recognition schemes usually show vulnerability under changing facial expressions. The proposed scheme discriminates input face images into one of several context categories. The context categories are decided by unsupervised learning method based on the facial geometries that are derived from either scanned mosaic face images and/or coordinates of facial feature points. The proposed method provides a self-adaptive preprocessing and feature representation in accordance with the identified context category using the genetic algorithm and knowledge accumulation mechanism. The superiority of the proposed method is shown using FERET database where face images are relatively exposed to wide range of facial expression variation.