Author(s)

Ai Le, Zhizhong Zhang

Corresponding Author

Ai Le

Abstract

In today's society, the growing instability has made identification technology a hot issue. Most traditional methods are based on a series of hand-defined preprocesses such as binarization. However, these preprocesses require strong prior knowledge. And that will lead to dropped or false extractions of minutiae. In this paper, a fingerprint triangle region recognition approach based on deep convolutional neural networks is proposed, which directly extract minutiae on raw fingerprint images without any preprocess since we tactfully take the advantage of the strong representative capacity of deep convolutional neural networks (including JudgeNet and LocateNet). Results show that our approach performs better in accuracy and robustness.

Keywords

Neural network, minutiae, fingerprint image, convolutional layer