Binary feature vector fingerprint representation from minutiae vicinities

Today, when comparing classical fingerprint matching and more constrained algorithms, like after binary quantization for biometric privacy protection purpose, there is an important gap in terms of performances. Performances of the latter solutions still need to be improved to decrease this gap. The main difficulty comes from the fact that fingerprint captures of the same trait give very different minutiae sets with possibly small overlaps and very different distortions among the different part of the images; and comparison of a stored reference with the fresh captured fingerprint data has to take into account those local variabilities. In this paper, we study a new approach to this problem by exhibiting a way to transform a minutiae set into a quantized feature vector by local comparisons. The encoding of the original fingerprint template is made by matching small minutiae vicinities with a set of representative vicinities. Moreover, the representation achieves the interesting property of self-alignment of the vectors.