Cancelable Biometric Recognition Using Deep Learning Based ResNet50 Model *

Biometrics are currently utilised to authenticate users of any organisation in order to offer security within that organisation. However, traditional biometric based system suffered from security and privacy issues. Cancelable biometric technique is used to transform the original biometric credentials i.e. face, ear, iris, fingerprint and fingervein etc. into distorted form which are non-invertible in nature. In this work, to enable cancelable biometric based authentication, deep learning based ResNet50 model has been used. Firstly, biometric features are extracted using ResNet50 model. Further, Gaussian Random Projection is used for generating the cancelable biometric template from extracted features. The classification accuracy is further computed using Random Forest Classifiers. Extensive experiments have been carried out on Fingerprint Verification Challenge 2002 (FVC2002), ORL and UTIRIS gray datasets while IIT Delhi iris and AMI ear datasets have been used for color datasets. On IIT Delhi iris and AMI ear color datasets, the accuracy achieved is 100% and 100% respectively while on ORL and FVC2002 the achieved accuracy is 96.25% and 96.88%. Further, the proposed work has also been compared with state-of-the-art methods which shows the outperformance of the proposed work over others.