Patch-Discontinuity Mining for Generalized Deepfake Detection

The rapid advancement of generative artificial intelligence has enabled the creation of highly realistic fake facial images, posing serious threats to personal privacy and the integrity of online information. Existing deepfake detection methods often rely on handcrafted forensic cues and complex architectures, achieving strong performance in intra-domain settings but suffering significant degradation when confronted with unseen forgery patterns. In this paper, we propose GenDF, a simple yet effective framework that transfers a powerful large-scale vision model to the deepfake detection task with a compact and neat network design. GenDF incorporates deepfake-specific representation learning to capture discriminative patterns between real and fake facial images, feature space redistribution to mitigate distribution mismatch, and a classification-invariant feature augmentation strategy to enhance generalization without introducing additional trainable parameters. Extensive experiments demonstrate that GenDF achieves state-of-the-art generalization performance in cross-domain and cross-manipulation settings while requiring only 0.28M trainable parameters, validating the effectiveness and efficiency of the proposed framework.

Key Contributions

• Deepfake-specific representation learning (DSRL) that captures patch-discontinuity patterns unique to AI-generated fake facial images vs. real ones in a low-dimensional space
• Feature space redistribution (FSR) to mitigate distribution mismatch between generic pre-trained vision models and the real/fake facial image domain
• Classification-invariant feature augmentation (CIFAug) that expands feature space orthogonally to the decision boundary, improving generalization to unseen forgery patterns without adding trainable parameters

🛡️ Threat Analysis

Details

Similar Papers