Exploring the Performance of Deterministic and Machine Learning Algorithms for Masking of Fingerprints on Public Documents

Fingerprints are often used as a unique identifier of a citizen on national identity cards (NICs), such as the Aadhar card in India. However, creating a NIC for everyone in the population, in a country like India, takes several years. However, fingerprints can also be accessed from public documents such as land deeds, especially for people who have not received formal education and hence prefer to use fingerprints instead of signatures. Such fingerprints can be used to create fake NICs with criminal intentions. Moreover, when someone’s fingerprints have already been used to create an NIC, they may face issues while trying to create an NIC for themselves. Fingerprint closing has been used in India to exploit AePS or the Adhaar enabled Payment System. Therefore, it is important to identify and anonymize all the fingerprints present in public documents. In this work, we explore two approaches, to segment fingerprints in government documents, one with a deterministic algorithms and another using traditional Machine Learning (ML) algorithms. We obtain an accuracy of 95.5% with the deterministic algorithms and 98.7% with the ML model.