A deep learning model for fingerprint biometric ıdentification

This thesis concerns the development, training, and testing of a deep learning model that recognizes fingerprints. We will use convolutional neural networks (CNNs) to carry out identification and verification activities. The study uses two different fingerprint sets containing various types of fingerprints, including contact-based and contactless images. These sets include the FVC2006 (DB1, DB2, DB3, and DB4) and the Hong Kong Polytechnic University Fingerprint Images Database. The study involves improving fingerprint images, extracting features from those images, and training a CNN classifier that can effectively handle the two different modalities. First, the FVC2006 DB1 Electric-Field dataset is investigated to study contact-based fingerprint recognition. The second stage considers the PolyU 2D-to-contact dataset, which consists of contact-based and contactless fingerprint images. Discriminative features with the potential for accurate fingerprint matching and classification are extracted using effective methods such as Gabor filters, orientation analysis, and texture descriptors. The single-point detection that was also studied includes core and delta detection, which is important for reliable fingerprint classification and matching. Some of the metrics used to measure the performance of the proposed system are classification accuracy, equal error rate (EER), and receiver operating characteristic (ROC) curves. The system attained classification accuracies of 92.76% on the v FVC2006 (DB1) dataset and 93.75% on the PolyU dataset. The EER values for the FVC2006 and PolyU datasets were 14.26% and 2.99%, respectively, demonstrating the efficiency of the CNN method for fingerprint recognition. This work contributes to the growing field of fingerprint biometrics by providing information about issues and processes in cross-modality fingerprint recognition. The results demonstrate the advantages of CNNs in enhancing the performance of fingerprint classification systems, particularly with regard to different acquisition methods and datasets.