Age and Spoof Detection from Fingerprints using Transfer Learning
Sonal Ayyappan, Navaneeth Asok, Nandu Shaji · 2021
The use of biometric authentication has seen an exponential increase in recent years ranging from smartphones to even forensic analyses. Fingerprints are obtained and used in crime scenes, old monuments and excavated relics and to the day-to-day authentication including attendance marking. Determination of age has always been an ardent task as they experience virtually zero changes as a person ages. Also, with the increased attacks and bypassing on the fingerprint authentication systems, it is also important to confirm the genuineness of the fingerprints. This brings forth a need for a spoof detection for fingerprints. Since fingerprints have been used as an effective method for authentication, their correlation with the age of a person is to be identified, if any. This paper aims in using Convolutional Neural Networks and other machine learning techniques to estimate age of a person from fingerprints and also spoof detection. The models we compare include three pre-trained CNNs which are fine-tuned with the fingerprint images, and a classical Local Binary Pattern approach. It is found that pre-trained CNNs along with Dataset Augmentation can produce good results with no need for any hyperparameter selection. NIST dataset was used for age detection and LiveDet 2013 dataset was used for spoof detection. It was able to achieve a top accuracy of 84% for age detection and 94% for spoof detection. The paper also focuses on identifying the best scanner for our purposes and also the possible materials used for spoofing.