Supervised generalized canonical correlation analysis of soft biometric fusion for recognition at a distance
B.H. Guo, Mark S. Nixon, J.N. Carter · 2017
In order to improve biometric system performance, information fusion becomes a key technique in multi-modal biometric systems. Multi-modal biometric fusion is conventionally divided into four levels: sensor level, feature level, score level and decision level. In this paper, we propose a supervised generalized canonical correlation (sg-CCA) method to fuse soft biometric features. The experiments were performed using a soft biometric database which contains the human face, body and clothing traits at three different distances. This paper describes the database and analyses the recognition performance. Furthermore, it explores the potency of face, body and clothing for human recognition using sg-CCA fusion compared with other linear dimensionality reduction fusion methods. The results demonstrate the superiority of soft biometric fusion using sg-CCA method for human recognition.