Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Unsupervised Fingerphoto Presentation Attack Detection With Diffusion Models

Hailin Li, Raghavendra Ramachandra, Mohamed Ragab, Soumik Mondal, Yong Kiam Tan, Khin Mi Mi Aung

TL;DR

This work tackles fingerphoto presentation attack detection under limited attack data by framing PAD as a one-class unsupervised problem. It trains a Denoising Diffusion Probabilistic Model (DDPM) on bona fide fingerphotos to reconstruct input ROI images and uses Learned Perceptual Image Patch Similarity (LPIPS) to detect attacks via reconstruction differences, guided by a DifFace-inspired posterior for HQ restoration. Across Clarkson, NTNU, and HDA datasets, the DDPM+LPIPS approach achieves superior detection performance compared to baselines, especially for challenging PAIs, though cross-dataset generalization remains sensitive to capture conditions. The study also analyzes the impact of ROI processing and metric choice, highlighting practical considerations for deploying diffusion-based fingerphoto PAD in real-world settings and suggesting avenues for dual use in image enhancement and de-occlusion. The proposed method offers a promising unsupervised route to robust fingerphoto PAD with potential for scalable defense against unseen PAIs.

Abstract

Smartphone-based contactless fingerphoto authentication has become a reliable alternative to traditional contact-based fingerprint biometric systems owing to rapid advances in smartphone camera technology. Despite its convenience, fingerprint authentication through fingerphotos is more vulnerable to presentation attacks, which has motivated recent research efforts towards developing fingerphoto Presentation Attack Detection (PAD) techniques. However, prior PAD approaches utilized supervised learning methods that require labeled training data for both bona fide and attack samples. This can suffer from two key issues, namely (i) generalization:the detection of novel presentation attack instruments (PAIs) unseen in the training data, and (ii) scalability:the collection of a large dataset of attack samples using different PAIs. To address these challenges, we propose a novel unsupervised approach based on a state-of-the-art deep-learning-based diffusion model, the Denoising Diffusion Probabilistic Model (DDPM), which is trained solely on bona fide samples. The proposed approach detects Presentation Attacks (PA) by calculating the reconstruction similarity between the input and output pairs of the DDPM. We present extensive experiments across three PAI datasets to test the accuracy and generalization capability of our approach. The results show that the proposed DDPM-based PAD method achieves significantly better detection error rates on several PAI classes compared to other baseline unsupervised approaches.

Unsupervised Fingerphoto Presentation Attack Detection With Diffusion Models

TL;DR

This work tackles fingerphoto presentation attack detection under limited attack data by framing PAD as a one-class unsupervised problem. It trains a Denoising Diffusion Probabilistic Model (DDPM) on bona fide fingerphotos to reconstruct input ROI images and uses Learned Perceptual Image Patch Similarity (LPIPS) to detect attacks via reconstruction differences, guided by a DifFace-inspired posterior for HQ restoration. Across Clarkson, NTNU, and HDA datasets, the DDPM+LPIPS approach achieves superior detection performance compared to baselines, especially for challenging PAIs, though cross-dataset generalization remains sensitive to capture conditions. The study also analyzes the impact of ROI processing and metric choice, highlighting practical considerations for deploying diffusion-based fingerphoto PAD in real-world settings and suggesting avenues for dual use in image enhancement and de-occlusion. The proposed method offers a promising unsupervised route to robust fingerphoto PAD with potential for scalable defense against unseen PAIs.

Abstract

Smartphone-based contactless fingerphoto authentication has become a reliable alternative to traditional contact-based fingerprint biometric systems owing to rapid advances in smartphone camera technology. Despite its convenience, fingerprint authentication through fingerphotos is more vulnerable to presentation attacks, which has motivated recent research efforts towards developing fingerphoto Presentation Attack Detection (PAD) techniques. However, prior PAD approaches utilized supervised learning methods that require labeled training data for both bona fide and attack samples. This can suffer from two key issues, namely (i) generalization:the detection of novel presentation attack instruments (PAIs) unseen in the training data, and (ii) scalability:the collection of a large dataset of attack samples using different PAIs. To address these challenges, we propose a novel unsupervised approach based on a state-of-the-art deep-learning-based diffusion model, the Denoising Diffusion Probabilistic Model (DDPM), which is trained solely on bona fide samples. The proposed approach detects Presentation Attacks (PA) by calculating the reconstruction similarity between the input and output pairs of the DDPM. We present extensive experiments across three PAI datasets to test the accuracy and generalization capability of our approach. The results show that the proposed DDPM-based PAD method achieves significantly better detection error rates on several PAI classes compared to other baseline unsupervised approaches.
Paper Structure (19 sections, 5 equations, 4 figures, 10 tables)

This paper contains 19 sections, 5 equations, 4 figures, 10 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Proposed Method
  4. Denoising Diffusion Probabilistic Model
  5. Learned Perceptual Image Patch Similarity
  6. Experimental Evaluation
  7. Datasets
  8. Evaluation Metric
  9. Performance Evaluation Protocol
  10. ROI Extraction.
  11. Baseline Unsupervised Models.
  12. Dataset Partition.
  13. Experiments.
  14. Result and Discussion
  15. Choice of Image Similarity Metric
  16. ...and 4 more sections

Figures (4)

  • Figure 1: Fingerphoto reconstructions using various unsupervised models from an input image (top-to-bottom): bona fide fingerphoto, then PAI attack images using Ecoflex, photopaper, Playdoh, and woodglue; models (left-to-right): convolutional autoencoder (CAE), variational autoencoder (VAE), StyleGAN-ADA (SGA), and DDPM; the input images are not seen during model training.
  • Figure 2: The pipeline of our proposed DDPM and LPIPS fingerphoto PAD method.
  • Figure 3: PAIs and bona fide samples from the three datasets.
  • Figure 4: Misclassification cases for DDPM input/output pairs. The first row shows two attack examples and the second row shows two bona fide examples.