TipSegNet: Fingertip Segmentation in Contactless Fingerprint Imaging

TL;DR

This work tackles fingertip segmentation in contactless fingerprint imaging by proposing TipSegNet, a direct hand-image segmentation model that bypasses separate finger-detection steps. The architecture combines a ResNeXt-101 backbone with a Feature Pyramid Network-inspired decoder, trained with an extensive data-augmentation regime to achieve state-of-the-art performance on multi-finger segmentation. Key results include a mean IoU of $0.987$ and an accuracy of $0.999$, with a high F1 score of $0.994$, demonstrating robustness across varied poses and backgrounds. The study also provides ablation insights on data augmentation and backbone choices, discusses practical deployment considerations (notably model size and computational cost), and outlines future directions such as model compression and integration with advanced biometric techniques to further improve real-world applicability.

Abstract

Contactless fingerprint recognition systems offer a hygienic, user-friendly, and efficient alternative to traditional contact-based methods. However, their accuracy heavily relies on precise fingertip detection and segmentation, particularly under challenging background conditions. This paper introduces TipSegNet, a novel deep learning model that achieves state-of-the-art performance in segmenting fingertips directly from grayscale hand images. TipSegNet leverages a ResNeXt-101 backbone for robust feature extraction, combined with a Feature Pyramid Network (FPN) for multi-scale representation, enabling accurate segmentation across varying finger poses and image qualities. Furthermore, we employ an extensive data augmentation strategy to enhance the model's generalizability and robustness. TipSegNet outperforms existing methods, achieving a mean Intersection over Union (mIoU) of 0.987 and an accuracy of 0.999, representing a significant advancement in contactless fingerprint segmentation. This enhanced accuracy has the potential to substantially improve the reliability and effectiveness of contactless biometric systems in real-world applications.

Figures (4)

• Figure 1: Examples from the training set with added augmentations. Input images as shown to the model during training on the left and their corresponding labels on the right.
• Figure 2: Model architecture. ResNeXt part is encircled by the dashed, green line and the output of its four main layers is used by the FPN to generate the multi-scale predictions (yellow), which are then upscaled (checkerboard pattern), before being summed together to create the input to the segmentation head (red).
• Figure 3: Training and validation loss over training epochs.
• Figure 4: Four exemplary segmentation results from the validation set. Class 0 describes the separation of the fingers from the background, class 1 the left index finger, class 2 the left middle finger and so on.