Fingerprint indexing schemes – A survey

Abstract

Biometric authentication involves identification/verification of an individual based on his physiological or behavioral characteristics. Fingerprint, a physiological characteristic, is an impression developed on a surface touched by a human fingertip. As the use of fingerprint recognition grows, more and more people get registered to the system and therefore the size of biometric database increases. During identification when a new fingerprint is provided and the task is to find the most similar fingerprint from the database, it essentially involves searching the entire database. Identification becomes very compute intensive for large databases. If we can somehow reduce the search space from the entire database to a small size list, the identification would become cost-effective. One possibility is to use locality sensitive hashing based approaches. The key idea is to choose an appropriate representation of the fingerprint and to device an indexing approach around it. Such a technique would, instead of directly comparing the query fingerprint with all the fingerprints stored in the database, make a preprocessing to produce a small size candidate list of fingerprints and then search within the list. The candidate list contains the target fingerprint with certain probabilistic guarantees. Indexing is challenging for biometric databases due to various factors such as the presence of the variable number of features, absence of structural information like order, the presence of occlusion and illumination in the image etc. This paper presents a survey of indexing schemes used for fingerprint databases. These schemes have been classified into four broad categories, namely texture-based, minutiae-based, hybrid and deep learning based schemes. The schemes broadly cover the use of Level-1, Level-2, texture, and deep features.