This repository contains the official source code for DB-FIQA, a fingerprint quality assessment framework designed for crime scene (latent) fingerprint analysis. The proposed method and benchmark have been accepted by IEEE Transactions on Image Processing (TIP).
Paper: Latent Fingerprint Quality Assessment for Criminal Investigations: A Benchmark Dataset and Method
Authors: Chao Huang, Jingxuan Zhang, Ye Zhang, Hao Wu, Peibei Cao, Zhihua Wang, Yang Yu, Xiaochun Cao
- Python 3.6+
- PyTorch 2.3.0
- TorchVision
Predicting fingerprint quality with our model trained on the CSFD-10k Dataset.
To run the demo, please download the pre-trained model at BaiduCloud[https://pan.baidu.com/s/1aedadrmb5PbTWL7-pFvTqg?pwd=d94v] password: d94v, put it in 'model/checkpoints/' folder, then run:
python predict_ones.py
A group of quality score ranging from 0-100 (a higher value indicates better image quality) will be stored in 'output_scores' directory.
Training and testing our model on the Crime Scene Fingerprints quality Dataset. The dataset images can be downloaded from BaiduCloud[https://pan.baidu.com/s/12j7_Yz_E9PhmibcLtAwYPw?pwd=2djs] password: 2djs. Alternative link: https://www.kaggle.com/datasets/zhangjingxuanddd/csfd-images. put the 'data' and 'data_e' folder into 'dataset' directory. corresponding image names and MOS are contained in this repository.
python train.py
Some available options:
--epochs: total epochs.--lr: learning rate.--batch_size: Batch size.--train_rate: The portion of training set in the whole dataset.
@article{huang2026dbfqa,
title = {Latent Fingerprint Quality Assessment for Criminal Investigations: A Benchmark Dataset and Method},
author = {Huang, Chao and Zhang, Jingxuan and Zhang, Ye and Wu, Hao and Cao, Peibei and Wang, Zhihua and Yu, Yang and Cao, Xiaochun},
booktitle = {IEEE Transaction on Image Processing},
year = {2026}
}