Ordinal representations for biometrics recognition

Biometrics provides a reliable method for automatic personal identification and has wide and important applications. The success of a biometric recognition system depends critically on its feature representation model for biometric patterns. The most challenging issue in biometric feature representation is to achieve sensitivity to inter-class differences and at the same time robustness against intra-class variations. Many biometric representation schemes have been reported but the above issue remains to be resolved. In this paper, we introduce ordinal measures for iris, face and palmprint image representation in an attempt to resolve this issue. With this so-called ordinal representation model in place, many best-performing biometric recognition methods may be interpreted as special cases of this model. In this sense, the proposed ordinal representation model forms a general framework for biometric pattern representation. Extensive experimental results on public biometric databases demonstrate the effectiveness and generality of this representation.