Article: Performance analysis on fingerprint identification by deep learning approach Journal: International Journal of Biometrics (IJBM) 2025 Vol.17 No.4 pp.406 - 431 Abstract: Achieving high accuracy in fingerprint identification remains challenging, despite various approaches that have been introduced over the years, including deep learning-based methods. These approaches can be computationally complex and may require a vast amount of training data. This study aims to evaluate the performance of deep learning-based approaches for fingerprint identification using two pretrained deep network models, <i>i.e.</i>, GoogLeNet and ResNet18. The images in the datasets are first registered and cropped before being trained and validated. The validation rates demonstrated that the preprocessed images produced higher average validation rates compared to the original images. These images are then applied during the testing phase, resulting in nearly perfect identification rates for both models. In comparison, with only 20% of the training dataset, GoogLeNet and ResNet18 achieved 93.00% and 97.00% for the FingerDOS database, respectively. Both models obtained an 88.75% identification rate on the FVC2002 DB1A database, outperforming other methods. Inderscience Publishers - linking academia, business and industry through research

Title: Performance analysis on fingerprint identification by deep learning approach

Authors: Florence Francis-Lothai; Kung Chuang Ting; Emily Sing Kiang Siew; Hai Inn Ho; Annie Joseph; Tengku Mohd Afendi Zulcaffle; David B.L. Bong

Addresses: Faculty of Engineering, Universiti Malaysia Sarawak, Sarawak, Malaysia; Faculty of Computing and Software Engineering, i-CATS University College, Sarawak, Malaysia ' Faculty of Computing and Software Engineering, i-CATS University College, Sarawak, Malaysia ' Faculty of Computing and Software Engineering, i-CATS University College, Sarawak, Malaysia ' Faculty of Computing and Software Engineering, i-CATS University College, Sarawak, Malaysia ' Faculty of Engineering, Universiti Malaysia Sarawak, Sarawak, Malaysia ' Faculty of Engineering, Universiti Malaysia Sarawak, Sarawak, Malaysia ' Faculty of Engineering, Universiti Malaysia Sarawak, Sarawak, Malaysia

Abstract: Achieving high accuracy in fingerprint identification remains challenging, despite various approaches that have been introduced over the years, including deep learning-based methods. These approaches can be computationally complex and may require a vast amount of training data. This study aims to evaluate the performance of deep learning-based approaches for fingerprint identification using two pretrained deep network models, i.e., GoogLeNet and ResNet18. The images in the datasets are first registered and cropped before being trained and validated. The validation rates demonstrated that the preprocessed images produced higher average validation rates compared to the original images. These images are then applied during the testing phase, resulting in nearly perfect identification rates for both models. In comparison, with only 20% of the training dataset, GoogLeNet and ResNet18 achieved 93.00% and 97.00% for the FingerDOS database, respectively. Both models obtained an 88.75% identification rate on the FVC2002 DB1A database, outperforming other methods.

Keywords: fingerprint identification; biometric; deep learning; GoogLeNet; ResNet18; image registration; speeded up robust features; SURF.

DOI: 10.1504/IJBM.2025.147210

International Journal of Biometrics, 2025 Vol.17 No.4, pp.406 - 431

Received: 21 Aug 2023
Accepted: 25 Aug 2024
Published online: 11 Jul 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: Performance analysis on fingerprint identification by deep learning approach

Keep up-to-date