Overlapped Fingerprint Separation Based on Deep Learning

Abstract

Biometrics and artificial intelligence play the important roles of recent technology. In biometrics, fingerprint is one of the most widely used identification methods. However, most of this kind applications only focus on single fingerprint processing but lack discussion of recognition of overlapped fingerprint due to its complexity. In fact, overlapped fingerprints are much more common on the criminal spot and nowadays we still rely on the inefficient manual operation to separate those overlapped fingerprints. So, we purpose our automatic, accurate, and even more efficient method using convolutional neural network to deal with the overlapped fingerprints problem. In experimental result, not only the single and multi-fingerprint latent test has 92.39% and 97.1% average accurate rate respectively, but we also got 92.19% and 95.84% correct rate respectively in the overlapped and non-overlapped range detection tests. The result shows that we could actually assist the fingerprint separation work automatically and efficiently with our own method.