Selection of Distinguish Points for Class Distribution Preserving Transform for Biometric Template Protection

This paper addresses the biometric template security issue. Follow out previous work on class distribution transform, the proposed scheme selects the distinguish points automatically. By considering the geometric relationship with the biometric templates, the proposed scheme transforms a real-value biometric template to a binary string such that the class distribution is preserved and proved mathematically. The binary string is then further encoded using BCH and hashing method to ensure that the template protecting algorithm is non-invertible. Two face databases, namely ORL and FERET, are selected for evaluation and LDA is used for creating the original template. Experimental results show that by integrating the proposed scheme into the LDA (original) algorithm, the system performance can be further improved by 1.1% and 4%, in terms of equal error rate, on ORL and FERET databases respectively. The results show that the proposed scheme not only can preserve the original template discriminant power, but also improve the performance if the original template is not fully optimized.