Level 3 Feature Based Fingerprint Identification

Abstract

In this thesis, two novel schemes have been proposed: one scheme on dots and incipient ridges extraction and another scheme on matching using level 2 and level 3 features. Dots and incipient ridges are extracted by tracing valley. Starting points are found on the valley by analyzing the frequencies present in the fingerprint. Valleys are traced from the starting point using Fast Marching Method (FMM). An intensity based checking method is used for finding these feature points. Delaunay triangle has been constructed using level 2 feature. A novel algorithm of selecting compatible triangle pair from Delaunay triangle is proposed. A novel set of feature parameters are constructed by establishing spatial relation between minutiae and dots-and-incipient.
Pore based matching has been performed using Robust Affine Iterative Closest Point algorithm. These extended features (dots, incipient ridges, and pores) are helpful for forensic experts. However, forensic experts deal with full-to-partial print matching of latent fingerprint. Hence, Full-to-partial fingerprint matching has been carried out. Partial print is constructed by cropping a window from a full fingerprint in two ways such as, non-overlapped cropping and random cropping. Form the experiment, it has been observed that random cropping based fingerprint has better accuracy than non-overlapped cropping. For performance evaluation of the proposed algorithm, two public databases have been used: NIST SD30 database and IIIT Delhi rural database. All images in SD30 are taken in constrained environment and images in IIIT database are taken in unconstrained environment. Feature level and score level fusion have been carried out for fusing different levels of feature. It has been observed that score level fusion shows better accuracy than feature level fusion.