Thinking in Frequency: Face Forgery Detection by Mining Frequency-aware Clues
Yu-Yang Qian, Guojun Yin, Lu Sheng, Zixuan Chen, Jing Shao · 2020 · 42 citationsRead the paper
As realistic facial manipulation technologies have achieved remarkable progress, social concerns about potential malicious abuse of these technologies bring out an emerging research topic of face forgery detection. However, it is extremely challenging since recent advances are able to forge faces beyond the perception ability of human eyes, especially in compressed images and videos. We find that mining forgery patterns with the awareness of frequency could be a cure, as frequency provides a complementary viewpoint where either subtle forgery artifacts or compression errors could be well described. To introduce frequency into the face forgery detection, we propose a novel Frequency in Face Forgery Network (F3-Net), taking advantages of two different but complementary frequency-aware clues, 1) frequency-aware decomposed image components, and 2) local frequency statistics, to deeply mine the forgery patterns via our two-stream collaborative learning framework. We apply DCT as the applied frequency-domain transformation. Through comprehensive studies, we show that the proposed F3-Net significantly outperforms competing state-of-the-art methods on all compression qualities in the challenging FaceForensics++ dataset, especially wins a big lead upon low-quality media.
What the full text shows
Working prototype- F3-Net (Xception) outperforms state-of-the-art methods on low-quality (LQ) media, specifically beating Xception-PAFilters in accuracy. — 90.43% Acc and 0.933 AUC vs 87.16% Acc (FaceForensics++ dataset, LQ setting)
- F3-Net (Slowfast) achieves higher accuracy and AUC than the baseline Slowfast model on low-quality media. — 93.02% Acc and 0.958 AUC vs 90.53% Acc and 0.936 AUC (FaceForensics++ dataset, LQ setting)
- F3-Net shows significant improvement in detecting NeuralTextures (NT) forgeries compared to Xception. — 4.2% improvement in Acc score (FaceForensics++ dataset, LQ setting)
- F3-Net (Xception) achieves near-perfect accuracy on raw, uncompressed videos. — 99.95% Acc (FaceForensics++ dataset, RAW setting)
- Optimal performance for Local Frequency Statistics (LFS) is achieved with a specific window size. — AUC 0.889 at window size 10 (FaceForensics++ dataset, LQ task)
- Setup
- FaceForensics++ dataset containing 1,000 real videos (720 train, 140 val, 140 test) and 4,000 fake videos across four manipulation methods (DeepFakes, Face2Face, FaceSwap, NeuralTextures) at three quality levels (RAW, HQ, LQ).
- How
- The authors proposed F3-Net, a two-stream collaborative learning framework using Discrete Cosine Transform (DCT) to extract frequency-aware decomposed image components (FAD) and local frequency statistics (LFS). These streams are fused using a cross-attention MixBlock and implemented using Xception or Slowfast-R101 backbones.
- Limitations
- Performance on NeuralTextures is lower than other manipulation types, though improved over baselines.
Summarized by Gemma 4 from the open-access full text · 8 Oct 2026. Read it