A Finger Vein Template Protection Algorithm Based on Multi-level Feature Fusion

The existing finger vein template protection algorithms are difficult to make full use of pixel layer and feature layer elements, and are not robust to the change of feature contrast. We proposed a new vein feature template protection algorithm based on Multi-level Feature Fusion (MFF). By introducing random token and local permutation strategy, the pixel layer and feature layer elements are encrypted and fused in parallel to ensure the irreversibility and revocability of the template. Firstly, the orthogonal matrix is generated by inner product operation between the real-valued eigenvectors of pixel layer and random matrix vectors. Secondly, a Stepwise Decision Diagram (SDD) is used to generate the permutation seed, which converts the real-valued orthogonal matrix into a binary hash sequence. Then, the hash template is obtained by XOR operation between the pixel permutation result and the partial permutation hash sequence of the feature layer and stored in the database. We tested and validated the proposed algorithm with four current representative algorithms on three different publicly available finger vein datasets. Experimental results show that the algorithm proposed in this paper achieves excellent EER performance in both token security and leakage, satisfying the requirements of revocable template protection.