Group collaborative representation with L2 norm regularization in finger-knuckle-print recognition

Finger-knuckle-print (FKP), the wrinkle image from finger-back surface, is a new biometric characteristic. A least square problem with L2-norm penalty called collaborative representation based classification (CRC), which is a algorithm developed from sparse representation based classifier (SRC), can also get high recognition rate as SRC, but lower computational cost than SRC. In this paper we do FKP recognition in representation-based classifier framework, and notice that the constraints are put on the all coefficients indiscriminately, ignoring the group information of training samples. To make full use of group priori knowledge, we propose a classifier in the CRC framework with group information called group-CRC (GCRC), in which the correlation between the query sample and training samples from different classes is considered as distinguishing constraint strength. The higher accuracy and good efficiency of the propose method is shown in the experiments. ©, 2015, Binary Information Press. All right reserved.