Adaptive latent fingerprint image segmentation and matching using Chan-vese technique based on EDTV model

• Main Authors: Hilles, Shadi M.S., Liban, Abdilahi, Mahmoud Altrad, Abdullah, Miaikil, Othman A.M., Baker El-Ebiary, Yousef A., Contreras, Jennifer O., Hilles, Mohanad M.
• Format: text
• Published: Animo Repository 2021
• Subjects: Image segmentation; Computer Sciences
• Online Access: https://animorepository.dlsu.edu.ph/faculty_research/12718
• Institution: De La Salle University

Biometrics such as face, fingerprint, iris, voice and palm prints are the most widely used, and as well the fingerprints are one of the most frequently used biometrics to identify individuals and authenticate their identity. commonly categorized into three different categories which are rolled, plain and latent fingerprints. The reliability of image segmentation for latent fingerprint which is used in criminal issues still challenges, The difficulty of latent fingerprint image segmentation mainly lies in the poor quality of fingerprint patterns and the presence of the noise in the background, This research has investigated the fingerprint segmentation and matching based on EDTV and presented Chan-vese active contour segmentation technique, in addition, presented NIST SD27 for grayscale dataset of latent fingerprint which is standard by National Institute of Standard and Technology, where is dataset have varieties of fingerprint image samples, a total about 258 of latent fingerprint, those samples collected from crime scenes and matching fingerprint and shown the performance of matching accuracy ROC and CMC curves, To evaluate the performance of the matching ROC and CMC curves has been deployed, The area under curve (AUC) of the ROC of the good images performance is 72% with CMC rank1-idnetification of 42% and rank-20 identification of 79%. the result shows that the latent fingerprint method performance is better for good latent fingerprint images compare to bad and ugly images, while there is no much difference for bad and ugly image.