Fingerprint matching using a deep learning based approach

Abstract

As we move towards a technological driven era, the traditional methods of data or personnel verification are becoming redundant and easier to crack. Biometric authentication has emerged as a very promising technique as it uses features of human body which are unique to every individual. In an effort to adopt recent developments in Machine Learning (ML) and Natural Language Processing (NLP) and apply them to the domain of biometric verification, I propose a novel Vision Transformer (ViT) based Siamese Network (SN) framework for fingerprint match ing. Our primary focus is holistic and a end-to-end pipeline has been constructed and implemented using an ensemble of task-specific algorithms to procure the best possible result from the model. I have also endeavoured to identify specific problems on the application of ViT to our problem statement and introduced two major mod ifications, Shifted Patch Tokenization (SPT) and Localized Self Attention (LSA) to tackle those shortcomings effectively. I propose two variations for the model, namely Intermediate-Merge (IM) Siamese Network and Late Merge (LM) Siamese Network and test the performances on a fingerprint dataset from IIT Kanpur. Keywords: Fingerprint Matching, Vision Transformer, Siamese Networks, Deep Learning