Selecting Distinctive Features to Improve Performances of Multidimensional Fuzzy Vault Scheme

Fuzzy vault scheme is one of the most popular biometric cryptosystems. However, the scheme is designed for set differences while Euclidean distance is often used in biometric techniques. Multidimensional fuzzy vault scheme (MDFVS) is a modified version that can be easily implemented based on biometric feature data. In MDFVS, every point is a vector, and Euclidean distance measure is used for genuine points filtering. To get better performances, the step of feature selection in the MDFVS algorithms is very important and should be well designed. In this paper we propose applying recognition rate to measure discrimination of features and selecting strong distinctive features into genuine points. Some principles of selecting strong distinctive features to compose genuine points are discussed. An implementation of MDFVS with feature selection is also presented. Experimental results based on palmprint show that the proposed feature selection approach improves the performances of MDFVS.