Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Optimizing DINOv2 with Registers for Face Anti-Spoofing

Mika Feng, Pierre Gallin-Martel, Koichi Ito, Takafumi Aoki

TL;DR

This paper tackles spoofing attacks in face recognition by preemptively detecting minute live-vs-spoof differences using a DINOv2-based detector augmented with register tokens. The method preserves generalization by fine-tuning only the last encoder block and introducing four register tokens to stabilize attention and emphasize fine-grained cues. Empirical results on the FAS Workshop and SiW datasets show improved ACER, AUC, and ACC over a baseline, with robust performance against unknown spoofing attacks, albeit with some EER trade-offs due to class-imbalance. The work advances practical face anti-spoofing by combining register-enhanced ViT representations with targeted data augmentation and a careful training strategy, enabling more reliable pre-authentication screening in real-world systems.

Abstract

Face recognition systems are designed to be robust against variations in head pose, illumination, and image blur during capture. However, malicious actors can exploit these systems by presenting a face photo of a registered user, potentially bypassing the authentication process. Such spoofing attacks must be detected prior to face recognition. In this paper, we propose a DINOv2-based spoofing attack detection method to discern minute differences between live and spoofed face images. Specifically, we employ DINOv2 with registers to extract generalizable features and to suppress perturbations in the attention mechanism, which enables focused attention on essential and minute features. We demonstrate the effectiveness of the proposed method through experiments conducted on the dataset provided by ``The 6th Face Anti-Spoofing Workshop: Unified Physical-Digital Attacks Detection@ICCV2025'' and SiW dataset. The project page is available at: https://gsisaoki.github.io/FAS-DINOv2-ICCVW/ .

Optimizing DINOv2 with Registers for Face Anti-Spoofing

TL;DR

This paper tackles spoofing attacks in face recognition by preemptively detecting minute live-vs-spoof differences using a DINOv2-based detector augmented with register tokens. The method preserves generalization by fine-tuning only the last encoder block and introducing four register tokens to stabilize attention and emphasize fine-grained cues. Empirical results on the FAS Workshop and SiW datasets show improved ACER, AUC, and ACC over a baseline, with robust performance against unknown spoofing attacks, albeit with some EER trade-offs due to class-imbalance. The work advances practical face anti-spoofing by combining register-enhanced ViT representations with targeted data augmentation and a careful training strategy, enabling more reliable pre-authentication screening in real-world systems.

Abstract

Face recognition systems are designed to be robust against variations in head pose, illumination, and image blur during capture. However, malicious actors can exploit these systems by presenting a face photo of a registered user, potentially bypassing the authentication process. Such spoofing attacks must be detected prior to face recognition. In this paper, we propose a DINOv2-based spoofing attack detection method to discern minute differences between live and spoofed face images. Specifically, we employ DINOv2 with registers to extract generalizable features and to suppress perturbations in the attention mechanism, which enables focused attention on essential and minute features. We demonstrate the effectiveness of the proposed method through experiments conducted on the dataset provided by ``The 6th Face Anti-Spoofing Workshop: Unified Physical-Digital Attacks Detection@ICCV2025'' and SiW dataset. The project page is available at: https://gsisaoki.github.io/FAS-DINOv2-ICCVW/ .
Paper Structure (22 sections, 3 equations, 3 figures, 3 tables)

This paper contains 22 sections, 3 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Proposed Method
  4. Overview
  5. DINOv2 with Registers
  6. Training Strategy
  7. Experiments and Discussion
  8. Experiments on FAS Workshop Dataset
  9. Dataset
  10. Experimental Condition
  11. Evaluation Metrics
  12. Exp. (i): Unfreezing layers for DINOv2 with registers
  13. Exp. (i): Unfreezing layers for DINOv2 with registers
  14. Exp. (ii): Comparison between the Baseline Method and the Proposed Method
  15. Experiments on SiW Dataset
  16. ...and 7 more sections

Figures (3)

  • Figure 1: Overview of the proposed method.
  • Figure 2: Distribution of the classes in the datasets provided from "The 6th Face Anti-Spoofing Workshop: Unified Physical-Digital Attacks Detection@ICCV2025": (a) training dataset and (b) validation dataset.
  • Figure 3: Example of face images after applying FAS-Aug.