A Mechanism to Optimize Fingerprint Searching in Large Databases

Abstract

Modern ngerprint identi cation systems are typically associated with large ngerprint databases. When establishing the identity of a query ngerprint, system has to go through all ngerprint entries in the database. In such cases, the identi cation process typically has an unacceptable long response time and it can be accelerated by trimming the search space. This thesis proposes a new searching mechanism which optimizes the identi cation process by reducing the search space. Proposing search mechanism uses a score-based indexing methodology which relies on a small set of reference ngerprints to build an index for the ngerprint database. These reference ngerprints are selected by dividing the ngerprint database in to separate ngerprint classes. In the proposing indexing methodology, given ngerprint is matched against all reference ngerprints and match scores are obtained. Then, index code is computed by normalizing each match score value. In the retrieval, similar ngerprints are identi ed by measuring the Pearson's correlation coe cient between index codes. Finally, proposing search mechanism outputs a small number of ngerprints which can be matched further to establish the identity of the query ngerprint. Experiments were conducted with NIST SD-27A ngerprint database and proposing mechanism performed quite well. Utilizing any ngerprint classi er or any ngerprint matcher, less computational power and high tolerance against distorted or noisy ngerprints are some advantages of this mechanism.