A Convolutional Neural Network Model Robust To Distorted Fingerprints

摘要

The greatest challenge in fingerprint recognition is verifying distorted fingerprints. Distortion in fingerprints may arise from errors introduced while acquiring fingerprints, the nature of the fingerprints or from how there were deposited (in the case of latent fingerprints). In this paper, two convolutional neural networks were trained using different approaches. The first was trained in a regular pattern while the second was trained with an approach that minimizes errors that arise from verifying distorted fingerprints, and hence proposed for training models to be robust to distorted fingerprints. The trained models were evaluated on good and distorted data-sets. Results are modest and show better performance in the second model, compared to the first.