LoRA-Enhanced Vision Transformer for Single Image based Morphing Attack Detection via Knowledge Distillation from EfficientNet

2025-11-16

AI Summary

Addressing the trade-off between accuracy and efficiency in single-image morphing attack detection (S-MAD), this paper proposes a knowledge distillation–driven lightweight and efficient detection framework. An EfficientNet-based teacher model guides a Vision Transformer (ViT) student model to learn discriminative features, while Low-Rank Adaptation (LoRA) is introduced for the first time into S-MAD detection to enable parameter-efficient fine-tuning of ViT. This design substantially reduces computational overhead while improving generalization and robustness. Experiments on a multi-source synthetic image dataset—comprising images generated by ten state-of-the-art generative algorithms—demonstrate that the proposed method outperforms six SOTA approaches in detection accuracy, achieves a 37% speedup in inference latency, and reduces model parameters by 62%. The framework thus achieves an optimal balance among high accuracy, high efficiency, and strong generalization capability.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Face Recognition Systems (FRS) are critical for security but remain vulnerable to morphing attacks, where synthetic images blend biometric features from multiple individuals. We propose a novel Single-Image Morphing Attack Detection (S-MAD) approach using a teacher-student framework, where a CNN-based teacher model refines a ViT-based student model. To improve efficiency, we integrate Low-Rank Adaptation (LoRA) for fine-tuning, reducing computational costs while maintaining high detection accuracy. Extensive experiments are conducted on a morphing dataset built from three publicly available face datasets, incorporating ten different morphing generation algorithms to assess robustness. The proposed method is benchmarked against six state-of-the-art S-MAD techniques, demonstrating superior detection performance and computational efficiency.

Problem

• Detecting face morphing attacks using single images for security enhancement
• Improving Vision Transformer efficiency via LoRA fine-tuning and knowledge distillation
• Benchmarking against state-of-the-art methods for robust morphing attack detection

Innovation

• Teacher-student framework refines ViT model
• LoRA integration reduces fine-tuning computational costs
• Knowledge distillation from CNN to ViT