In this paper, we propose a method of fixed-length binary string extraction (denoted by LogGM_[)ROBA) from low-resolution palmprint image for developing palmprint template protection technology. In order to extract reliable (stable and discriminative) bits, multi-bit equal-probability-interval quantization and detection rate optimized bit allocation (DROBA) are operated on the real-valued features, which are resulted from representing the palmprint image by simple statistics on logarithmic transform of Gabor magnitude (LogGM). Assuming the Helper Data Scheme with a BCH error correction coding is adopted for template protection, the performance is evaluated on the Hong Kong PolyU palmprint database. The experimental results show that our method can achieve low Bit Error Rate (BER) resulted from genuine binary strings so that a long secret key (around 100 bits) is allowed to be combined for security, and low False Rejection Rate and low False Acceptance Rate (FRR/FAR) when the key retrial process is considered as a Hamming distance classifier, which verify the high stability and strong distinctive ability of our extracted palmprint binary string.
展开▼
