Towards Robust DeepFake Detection under Unstable Face Sequences: Adaptive Sparse Graph Embedding with Order-Free Representation and Explicit Laplacian Spectral Prior

View on arXiv ← Back to list

Authors: Chih-Chung Hsu, Shao-Ning Chen, Chia-Ming Lee, Yi-Fang Wang, Yi-Shiuan Chou

Published: 2025-12-08 12:31:07+00:00

AI Summary

The paper proposes a Laplacian-Regularized Graph Convolutional Network (LR-GCN) designed for robust DeepFake detection under unstable face sequences characterized by missing, shuffled, or corrupted frames. LR-GCN constructs an Order-Free Temporal Graph Embedding (OF-TGE) using semantic affinities and employs dual-level sparsity to isolate noise propagation. Crucially, it incorporates an explicit Graph Laplacian Spectral Prior to realize a task-driven spectral band-pass mechanism that highlights subtle forgery artifacts while suppressing random noise and background semantics.

Abstract

Ensuring the authenticity of video content remains challenging as DeepFake generation becomes increasingly realistic and robust against detection. Most existing detectors implicitly assume temporally consistent and clean facial sequences, an assumption that rarely holds in real-world scenarios where compression artifacts, occlusions, and adversarial attacks destabilize face detection and often lead to invalid or misdetected faces. To address these challenges, we propose a Laplacian-Regularized Graph Convolutional Network (LR-GCN) that robustly detects DeepFakes from noisy or unordered face sequences, while being trained only on clean facial data. Our method constructs an Order-Free Temporal Graph Embedding (OF-TGE) that organizes frame-wise CNN features into an adaptive sparse graph based on semantic affinities. Unlike traditional methods constrained by strict temporal continuity, OF-TGE captures intrinsic feature consistency across frames, making it resilient to shuffled, missing, or heavily corrupted inputs. We further impose a dual-level sparsity mechanism on both graph structure and node features to suppress the influence of invalid faces. Crucially, we introduce an explicit Graph Laplacian Spectral Prior that acts as a high-pass operator in the graph spectral domain, highlighting structural anomalies and forgery artifacts, which are then consolidated by a low-pass GCN aggregation. This sequential design effectively realizes a task-driven spectral band-pass mechanism that suppresses background information and random noise while preserving manipulation cues. Extensive experiments on FF++, Celeb-DFv2, and DFDC demonstrate that LR-GCN achieves state-of-the-art performance and significantly improved robustness under severe global and local disruptions, including missing faces, occlusions, and adversarially perturbed face detections.

Key findings

LR-GCN achieves state-of-the-art detection performance and demonstrates significantly improved robustness across all benchmark datasets, particularly under high levels of perturbation (masking ratios up to 0.8). For instance, LR-GCN maintained an AUC of 0.983 on FF++ at $mr=0.8$, far surpassing competing methods whose performance degraded substantially. Ablation studies confirm that the combination of the adaptive GCN, Graph Laplacian Spectral Prior (GLSP), and Sparsity Constraint (SC) is essential for achieving this high resilience to noisy inputs.

Approach

The method uses CNN features extracted from face regions to build an adaptive sparse graph (OF-TGE) based on feature affinities, independent of temporal order. A dual-level sparsity mechanism prunes both weak graph connections and noisy node features. Detection is performed by LR-GCN, which combines a Graph Laplacian high-pass pre-filter with GCN low-pass aggregation to selectively preserve mid-to-high frequency forgery cues.

Datasets

FF++, Celeb-DFv2, DFDC

Model(s)

Laplacian-Regularized Graph Convolutional Network (LR-GCN), CSPNet (as backbone)

Author countries

Taiwan

← Previous