sarwarmurshed / crfseg

Clarkson Rotated Fingerprint Segmentation (CRFSEG) system

This repository is associated with a research initiative exploring deep learning-based solutions for slap fingerprint segmentation. It contains the code for the CRFSEG (Clarkson Rotated Fingerprint Segmentation) model, specifically designed to segment slap fingerprint images. It also contains code to develop a transformer-based slap segmentation system (TransSEG) by leveraging FPNFormer.

Project Overview

Deep Age-Invariant Fingerprint Segmentation System.

In this research, we developed 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. Here is the complete architecture of the Clarkson Rotated Fingerprint Segmentation (CRFSEG) system.

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.

Key features

1. CRFSEG accurately segments slap fingerprint images, producing precise bounding boxes around each fingerprint and classifying them correctly.
2. CRFSEG is age and orientation invariant, capable of processing slap images from various age groups and orientations.
3. It is also easily deployable and adaptable to new datasets with minimal fine-tuning using a small number of images.

Contributions

1. Two new in-house large datasets named Combined and Challenging slap datasets contain 133,611 slap fingerprint images of children and adults.
2. Annotated all slap fingerprint images manually to establish a ground-truth baseline for the accuracy assessment of different fingerprint segmentation systems.
3. Developed a novel age-invariant deep learning-based slap segmentation model (viz. CRFSEG), that can handle arbitrarily oriented fingerprints of adult and juvenile subjects.
4. Built a transformer-based fingerprint segmentation model named TransSEG to compare CRFSEG with this state-of-the-art deep learning model.
5. Evaluated the performance of state-of-the-art commercial and non-commercial fingerprint segmentation systems named VeriFinger, and NFSEG on our noble dataset and compared the performance of these systems with the newly developed CRFSEG model and also compared with TransSEG.
6. Released trained CRFSEG and TransSEG models through GitHub for public use.

Data Availability Statement

The fingerprint dataset used in this research, collected under IRB-approved protocol, contains sensitive information about juvenile subjects and cannot be publicly shared due to legal and ethical constraints. However, to facilitate further research, interested parties can provide their code/models to the corresponding authors. The authors will evaluate the provided code/models on the slap fingerprint dataset and share the anonymized results with the requesting party.

Code details

• The crfseg folder contains resources related to the CRFSEG model
• The transSEG folder contains resources related to the TransSEG model
• We are currently only publishing code to run the CRFSEG and TransSEG models. We will publish the full code for training these models once our paper is accepted.

Citation

If you find this work useful in your research, please cite our paper:

@article{murshed2023deep,
  title={Deep Age-Invariant Fingerprint Segmentation System},
  author={Murshed, MG and Bahmani, Keivan and Schuckers, Stephanie and Hussain, Faraz},
  journal={arXiv preprint arXiv:2303.03341},
  year={2023}
}

crfseg's Issues

No folder" trained_model " was provided

Thanks for sharing your code on slap segmentation. I'm a student and I would like to test it. however no trained model was provided.
Could you release the trained model please?