A Novel Composite Framework for Large-Scale Fingerprint Database Indexing and Fast Retrieval

Abstract

This paper addresses the problem of fast fingerprint retrieval in a large database using clustering-based descriptors. Common solutions fall into two categories: classification and indexing. However, those methods only use one form for space narrowing and neglecting the complementary uniqueness of classification and indexing. In addition, there are two major problems in fingerprint identification: partialness and non-linear distortion. Recently, many proposed features focus on global and minutia information and both of them can not deal well with those two problems. This paper has three contributions. First, it proposes a composite classification-indexing-retrieval framework that greatly reduces time complexity. Second, clustering-based descriptors are extracted for indexing so that the search space is narrowed largely. Third, the class-jumping principle (CJP) is proposed to determine the correctness of classification and handle the problem of misclassification.