Fusion for Component Based Face Recognition

This paper proposes a practical way to realize the diversity in face recognition system for performance improvement by fusing the classification results from the components (characteristic regions such as eyes, nose and mouth) and from the whole face image, instead of concatenating the face feature and the modular features for a single classifier. The extracted sub-images are not totally independent from the face image, but the experiments show that the fused result is improved from the recognition result based on the face or components alone. The fusion is implemented and compared at both score level and decision level. Communication resources are preserved between the sensor and fusion point in decision level fusion at the expense of performance, and the selection of which fusion scheme to use depends on the system resources and performance requirement.