Binary palmprint representation for feature template protection

The major challenge of biometric template protection comes from the intraclass variations of biometric data. The helper data scheme aims to solve this problem by employing the Error Correction Codes (ECC). However, many reported biometric binary features from the same user reach bit error rate (BER) as high as 40%, which exceeds the error correcting capability of most ECC (less than 25%). Therefore, a novel palmprint binary feature extraction method is proposed in this paper. The real-valued features are firstly extracted. Then one-bit quantization and reliable bits selection are processed. For verification multiple samples are required to be enrolled while training is not necessary. Experiments have been carried out on the HongKong PolyU Palmprint database. Results show that our method achieves much lower BER, lower verification error rate and allows a secret key long enough for security.