A two -step hybrid approach for voiceprint-biometric template protection

摘要

Biometric template protection is a crucial issue to be addressed for widespread deployment of biometrics-based recognition systems in real life application. Although a number of biometric template protection methods have been reported, it is still a challenging task to devise a scheme to satisfy both security and performance. In this paper, a two-step hybrid approach is proposed to generate a cancelable voiceprint template utilizing the advantages of both Template Transformation and Biometric Cryptosystem. The original voiceprint is transformed with random matrix based on the similarity-preserving of random projection. Chaff points are added to the codebook and matching is also performed in the transformed domain. Random projection improves the cancelability while chaff points conceal the genuine codeword to enhance the security. Binary indexes help identify the genuine codeword accurately. The effectiveness of the proposed method is well supported by detailed analysis. The experimental results demonstrate that the recognition performance is well-kept as the original template does.