Fingerprint Classification Based on Deep Learning Approaches: Experimental Findings and Comparisons

Biometric classification plays a key role in fingerprint characterization, especially the identification process. In fact, reducing number of comparisons biometric recognition systems is essential when dealing with large-scale databases. The fingerprints aims to achieve this target by splitting into different categories. general approach requires pre-processing techniques that are usually computationally expensive. Deep Learning emerging as leading field has been successfully applied many areas, such image processing. This work shows performance pre-trained Convolutional Neural Networks (CNNs), tested on two databases—namely, PolyU and NIST—and other results presented literature order establish type allows us obtain best terms precision model efficiency, among approaches under examination, namely: AlexNet, GoogLeNet, ResNet. We present first study extensively compares most used CNN architectures classifying four, five, eight classes. From experimental results, was obtained database all due higher quality its samples. To confirm reliability our obtained, statistical analysis based McNemar test performed.