TipSegNet: Fingertip Segmentation in Contactless Fingerprint Imaging.

Area of Science:

• Computer Science
• Biometrics
• Image Processing

Background:

• Contactless fingerprint recognition offers advantages over traditional methods but requires precise fingertip segmentation.
• Accurate segmentation is challenging due to varying finger poses and background conditions.

Purpose of the Study:

• To introduce TipSegNet, a novel deep learning model for accurate fingertip segmentation in contactless biometrics.
• To enhance the performance and robustness of contactless fingerprint recognition systems.

Main Methods:

• Developed TipSegNet, a deep learning model utilizing a ResNeXt-101 backbone and Feature Pyramid Network (FPN).
• Employed extensive data augmentation for improved generalizability.
• Trained and evaluated the model on a dataset of 2257 labeled hand images.

Main Results:

• TipSegNet achieved state-of-the-art performance in fingertip segmentation.
• The model attained a mean intersection over union (mIoU) of 0.987 and an accuracy of 0.999.
• Demonstrated superior performance compared to existing methods.

Conclusions:

• TipSegNet represents a significant advancement in contactless fingerprint segmentation.
• The model's high accuracy can substantially improve the reliability of real-world contactless biometric systems.