Deep Age-Invariant Fingerprint Segmentation System

• Published in: IEEE transactions on biometrics, behavior, and identity science behavior, and identity science, 2024-11, p.1-1
• Main Authors: Murshed, M.G. Sarwar, Bahmani, Keivan, Schuckers, Stephanie, Hussain, Faraz
• Format: Article
• Language: English
• Subjects: Accuracy; Convolutional neural networks; deep slap Segmentation; Faster-RCNN; Fingerprint recognition; Image matching; Image segmentation; Juvenile and adult fingerprints; NFSEG; Shape; Skin; Software algorithms; Thumb; Transformer; Transformers; VeriFinger
• ISSN: 2637-6407
• DOI: 10.1109/TBIOM.2024.3506926

Fingerprints are an important biometric modality used in various applications, including border crossings, healthcare systems, criminal justice, electronic voting, and more. Fingerprint-based identification systems attain higher accuracy when utilizing a slap fingerprint image containing multiple fingerprints of a subject, as opposed to using a single fingerprint. However, segmenting or auto-localizing the fingerprints in a slap image is a challenging task due to factors such as the different orientations of fingerprints, noisy backgrounds, and the smaller size of fingertip components. Real-world slap image datasets often contain rotated fingerprints, making it challenging for biometric recognition systems to automatically localize and label them accurately. Errors in fingerprint localization and finger labeling lead to poor matching performance. In this paper, we introduce a deep learning-based method for generating arbitrarily angled bounding boxes to precisely localize and label fingerprints in both axis-aligned and over-rotated slap images. We present CRFSEG (Clarkson Rotated Fingerprint Segmentation Model), an improvement upon the Faster R-CNN algorithm, incorporating arbitrarily-angled bounding boxes for enhanced performance on challenging slap images. CRFSEG demonstrates consistent results across different age groups and effectively handles over-rotated slap images. We evaluated CRFSEG against the widely used slap segmentation systems NFSEG and VeriFinger. Additionally, we leveraged a transformer-based vision architecture to build TransSEG (Transformer-based Slap Segmentation System), a new model for further comparison of CRFSEG with state-of-the-art deep learning-based image segmentation models. In our dataset containing both normal and rotated images of adult and children subjects, CRFSEG achieved a matching accuracy of 97.17%, which outperformed TransSEG (94.96%), VeriFinger (94.25%) and NFSEG segmentation systems (80.58%). The results indicate that our novel deep learning-based slap segmentation system is more efficient for both children and adult slaps. The code for building the CRFSEG and TransSEG model is publicly available at https://github.com/sarwarmurshed/CRFSEG.