Fingerprint Alterations Type Detection and Gender Recognition Using Convolutional Neural Networks and Transfer Learning

Abstract

Biometrics are physical and behavioural characteristics that are unique to each individual and are used for digital identification and authentication of individuals. Fingerprints, a physiological biometric, are widely used by law enforcement and border control agencies for the identification of individuals. Criminals may purposefully alter their fingerprints with the intent of masking their identities to evade such agencies. Automatic Fingerprint Identification Systems (AFIS) traditionally do not have the capability of identifying fingerprint alterations; moreover, gender recognition of criminal offenders is vital as it significantly reduces investigation time, hence highlighting the necessity of the proposed research. This paper proposes the use of deep learning models to accurately identify real and altered fingerprints and moreover identify the type of alterations and recognise the gender of the individual. The most common type of alterations such as obliteration, z-cut, and central rotation alterations are mentioned in the Sokoto Coventry Fingerprint dataset which has been utilised in this paper. Convolutional neural network (CNN) and transfer learning architectures were used, out of which AlexNet achieved the highest classification accuracy of 98.50%, 94.84% and 83.07% on alteration detection, alteration type detection and gender classification, respectively.