Comprehensive fingerprint recognition utilizing one shot learning with Siamese Network

Abstract

Detailed Fingerprint investigation has been a dominant law enforcement tool which is utilized to distinguish suspects, settle crimes and violations for over 100 years. Moreover, gender classification from fingerprints is a vital step in forensic anthropol ogy in order to identify a criminal’s gender and reduce the list of suspects. A novel approach of machine learning (ML) which is One Shot Learning has been intro duced in this report for identification of persons which will implement the Siamese learning approach for training fingerprint samples by using the triplet loss. One Shot Learning has shown to be efficient because it reliably performs with only one labeled training example and one or a few training sets. Moreover, by using Transfer Learn ing with EfficientNetV2S an accuracy of 99.80%, 99.73%, 97.09%, 99.66%, 98.61% for identification of person, gender, hand, finger and detection of forge fingerprints has been achieved on the Sokoto Coventry Fingerprint Dataset.