A Real Helper Data Scheme

The helper data scheme utilizes a secret key to protect biometric templates. The current helper data scheme requires binary feature representations that introduce quantization error and thus reduce the capacity of biometric channels. For spectral-minutiae based fingerprint recognition systems, Shannon theory proves that the current helper data scheme cannot have more than 6 bits. A 6-bit secret key is too short to secure the storage of biometric templates. Therefore, we propose a new helper data scheme without quantization. A basic realization is to convert the real-valued feature vector into a phase vector. By applying the spectral minutiae method in the FVC2000-DB2 fingerprint database, our new helper data scheme together with repetition codes and BCH codes allows at least 76 secret bits.