LAKAN: Landmark-assisted Adaptive Kolmogorov-Arnold Network for Face Forgery Detection

2025-10-01

AI Summary

Existing CNN- and Transformer-based deepfake detection models exhibit limited capacity in modeling complex nonlinear artifacts inherent in forged facial images. To address this, we propose a novel detection framework leveraging the Kolmogorov–Arnold Network (KAN). Our method introduces two key innovations: (1) learnable spline-based activation functions to enhance local nonlinear approximation capability; and (2) geometric prior-guided parameter generation—where facial landmarks dynamically instantiate KAN module parameters, enabling structural awareness and adaptive feature modeling. Integrated into a generic image encoder architecture, the proposed approach achieves state-of-the-art performance on benchmark datasets including FaceForensics++ and Celeb-DF, improving average detection accuracy by 2.3–4.1 percentage points over prior methods. Moreover, it demonstrates superior cross-dataset generalization, confirming the effectiveness and robustness of KANs for deepfake detection.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

The rapid development of deepfake generation techniques necessitates robust face forgery detection algorithms. While methods based on Convolutional Neural Networks (CNNs) and Transformers are effective, there is still room for improvement in modeling the highly complex and non-linear nature of forgery artifacts. To address this issue, we propose a novel detection method based on the Kolmogorov-Arnold Network (KAN). By replacing fixed activation functions with learnable splines, our KAN-based approach is better suited to this challenge. Furthermore, to guide the network's focus towards critical facial areas, we introduce a Landmark-assisted Adaptive Kolmogorov-Arnold Network (LAKAN) module. This module uses facial landmarks as a structural prior to dynamically generate the internal parameters of the KAN, creating an instance-specific signal that steers a general-purpose image encoder towards the most informative facial regions with artifacts. This core innovation creates a powerful combination between geometric priors and the network's learning process. Extensive experiments on multiple public datasets show that our proposed method achieves superior performance.

Problem

• Detecting face forgeries in deepfake-generated media
• Modeling complex non-linear forgery artifacts effectively
• Guiding detection focus to critical facial regions

Innovation

• Uses learnable splines for flexible activation functions
• Integrates facial landmarks to guide network focus
• Dynamically generates KAN parameters using landmark priors