Fingerprint Pattern Identification and Classification
Xiaomeng Guo, Fan Wu, Xiaoyong Tang · 2018
Fingerprint as a unique feature of each person can be divided into different types, in this paper, we identify real fingerprints pattern and classify them with convolutional neural networks (CNN). The traditional fingerprint pattern classification method is only classified by artificially defined features, which is supervised learning, and requires much more time and is difficult to adapt to all the mass fingerprint database. Therefore, we propose an algorithm to recognize and classify the pattern directly on the fingerprint image. The new neural network (FCTP-Net) consists of four convolutional layers, three max-pooling layers and three fully-connected layers. By using the ability of automatic learning and feature extraction of convolutional neural network, we can obtain the pattern features from a large number of fingerprint data and compare with the previous experimental results to get the method of convergence during less time, and get accuracy enhancement of classification. The best training accuracy we get is 91.5 % of six-categories database.