DiffusionFF: Face Forgery Detection via Diffusion-based Artifact Localization

View on arXiv ← Back to list

Authors: Siran Peng, Haoyuan Zhang, Li Gao, Tianshuo Zhang, Bao Li, Zhen Lei

Published: 2025-08-03 18:06:04+00:00

AI Summary

DiffusionFF enhances face forgery detection by localizing artifacts using a denoising diffusion model. This model generates high-quality Structural Dissimilarity (DSSIM) maps, which are fused with features from a pre-trained forgery detector, improving detection accuracy and providing fine-grained localization.

Abstract

The rapid evolution of deepfake generation techniques demands robust and accurate face forgery detection algorithms. While determining whether an image has been manipulated remains essential, the ability to precisely localize forgery artifacts has become increasingly important for improving model explainability and fostering user trust. To address this challenge, we propose DiffusionFF, a novel framework that enhances face forgery detection through diffusion-based artifact localization. Our method utilizes a denoising diffusion model to generate high-quality Structural Dissimilarity (DSSIM) maps, which effectively capture subtle traces of manipulation. These DSSIM maps are then fused with high-level semantic features extracted by a pretrained forgery detector, leading to significant improvements in detection accuracy. Extensive experiments on both cross-dataset and intra-dataset benchmarks demonstrate that DiffusionFF not only achieves superior detection performance but also offers precise and fine-grained artifact localization, highlighting its overall effectiveness.

Key findings

DiffusionFF achieves state-of-the-art performance in cross-dataset and intra-dataset face forgery detection benchmarks. It surpasses existing methods in both detection accuracy and the precision of artifact localization, demonstrating its superior effectiveness and explainability. Ablation studies confirm the effectiveness of the proposed approach.

Approach

DiffusionFF uses a denoising diffusion model to generate high-quality DSSIM maps that highlight forgery artifacts. These maps are then fused with features from a pre-trained forgery detector to improve both detection accuracy and artifact localization.

Datasets

FaceForensics++, Celeb-DeepFake-v2, DeepFake Detection Challenge (DFDC), DeepFake Detection Challenge Preview (DFDCP), FFIW-10K

Model(s)

U-Net (denoising diffusion model), ConvNeXt-B (pre-trained forgery detector), EfficientNet-B4 (ablation study), Swin-B (ablation study)

Author countries

China

← Previous