Generalized Disguise Makeup Presentation Attack Detection Using an Attention-Guided Patch-Based Framework

2026-04-30 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy · Depth: Expert, extended

Summary

Researchers from Shahid Beheshti University propose a generalized framework for detecting disguise makeup presentation attacks, which are particularly challenging due to their realistic alteration of facial appearance using cosmetics and prosthetics. The framework employs a two-phase design: a style-invariant full-face model, trained with metric learning and enhanced by a whitening transformation, extracts region attention scores via Grad-CAM. These scores then guide a patch-based phase that performs localized analysis using region-specific subnetworks, also trained with metric learning, for fine-grained discrimination. The team also created a new, diverse dataset of live and disguise makeup faces under real-world conditions. Experimental results show strong generalization across both their collected dataset and SIW-Mv2, achieving 8.97% ACER and 9.76% EER on the collected dataset, and 0% ACER on Obfuscation and Impersonation attacks and 1.34% on Cosmetics attacks of SIW-Mv2, outperforming prior works.

Key takeaway

For research scientists developing robust face anti-spoofing systems, this framework demonstrates that combining global attention with localized, patch-based analysis and metric learning significantly improves detection of sophisticated disguise makeup attacks. You should consider integrating similar two-phase, attention-guided architectures and style-invariant feature extraction to enhance generalization across diverse and challenging presentation attack scenarios, especially those involving subtle or localized alterations.

Key insights

A two-phase, attention-guided, patch-based framework effectively detects realistic disguise makeup attacks.

Principles

• Combine global context with local detail.
• Metric learning improves embedding separability.
• Style invariance enhances generalization.

Method

A full-face network generates attention scores via Grad-CAM, guiding patch-specific subnetworks for localized feature learning and classification, both utilizing metric learning and style-invariant features.

In practice

• Use SPIGA for robust facial landmark detection.
• MobileNetV2 offers efficient feature extraction.
• Concatenate patch embeddings for final classification.

Topics

• Disguise Makeup Detection
• Face Anti-Spoofing
• Presentation Attack Detection
• Attention-Guided Learning
• Patch-Based Framework

Code references

• serengil/deepface
• andresprados/SPIGA

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• Multi-Frequency Fusion for Robust Video Face Forgery Detection — Lightweight fusion of handcrafted cues can significantly improve face forgery detection accuracy with minimal model overhead.
• MakeupMirror: Improving Facial Attribute Preservation in Diffusion Models for Makeup Transfer — MakeupMirror enhances diffusion-based makeup transfer by preserving facial identity and skin tone through specific architectural and sampling innovations.
• On the Impact of Face Segmentation-Based Background Removal on Recognition and Morphing Attack Detection — Face segmentation for background removal impacts both recognition accuracy and morphing attack detection in biometric systems.
• Where, What, Why: Toward Explainable 3D-GS Watermarking — A representation-native 3D-GS watermarking framework decouples carrier selection and visual compensation for robust, imperceptible, and explainable embedding.
• SEAL: Semantic-aware Single-image Sticker Personalization with a Large-scale Sticker-tag Dataset — SEAL improves single-image sticker personalization in diffusion models by mitigating overfitting and enhancing contextual control.
• Detecting Deepfakes with Multivariate Soft Blending and CLIP-based Image-Text Alignment — Multimodal image-text alignment and blended data augmentation significantly improve deepfake detection generalization and robustness.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.