Efficient and effective content-based image retrieval framework for fingerprint databases

Student: Javier MontoyaAdvisor: Prof. Dr. Neucimar J. LeiteCo-advisor: Prof. Dr. Ricardo TorresDegree: MasterAcademic Institution: Institute of Computing (UNICAMP)Enrollment’s date: 2005Expected graduation and dissertation’s date: March - 2007Completed activities: Completed credits, master’s qualification, implementatio n of tex-tural image descriptors.Abstract. Two kinds of fingerprint identification approaches have been pro-posed in the literature to reduce the number of one-to-many comparisons duringfingerprint image retrieval, namely, exclusive and continu ous classification. Al-though exclusive classification approaches reduce the numb er of comparisons,they present some shortcomings, including fingerprint ambi guous classification,and unbalanced fingerprint classificationdistribution. On the other side, contin-uous classification approaches have not been further studie d. In this context, wepropose an original continuous approachto guide the search and the retrieval infingerprint image databases considering both effectivenes s and retrieval speed.For that purposes, we use feature extraction and indexing methods consideringthe textural and directional information contained in finge rprint images. Pre-liminary results of our work involves a comparative study of several texturalimage descriptors obtained by combining different types of the Wavelet Trans-form with similarity measures. From our experiments we can conclude that thebest retrieval accuracy was achieved by combining Gabor Wavelets (GWs) withthe Square Chord similarity measure. Furthermore, the presence of noise anddistortions in fingerprint images have affected the overall retrieval accuracy.