A Proposed Siamese Convolutional Neural Network for Fingerprint Recognition

Abstract:
In the past, the main way for fingerprint recognition was based on traditional image processing technology, which is timeconsuming and easy to fail. In order to recognize fingerprints more efficiently, in this experiment, the study proposed to use a method based on Siamese Network consisting of Convolutional Neural Network (CNN), which was seldom studied in the past. This experiment also is employed the idea of Multi-Scale Dilated Convolutions which can improve the receptive field and make each output of the Convolution contains a large amount of information. Dilated Convolutions introduces a hyper-parameter called dilation rate, which defines the spacing of values when Convolutional kernels process the data. This model added the three convolutional layers together which dilation rate equals 1, 2, 3 respectively to get dilated convolutions. This study input fingerprints and measured the similarity via output distance. The smaller the distance, the more similar the fingerprint and therefore it can judge if the fingerprints belong to the same person. The experimental results support this view.