FreqBlender: Enhancing DeepFake Detection by Blending Frequency Knowledge

View on arXiv ← Back to list

Authors: Hanzhe Li, Jiaran Zhou, Yuezun Li, Baoyuan Wu, Bin Li, Junyu Dong

Published: 2024-04-22 04:41:42+00:00

AI Summary

FreqBlender enhances deepfake detection by generating pseudo-fake faces through frequency knowledge blending. It addresses limitations of spatial-domain blending methods by focusing on frequency distribution of forgery traces, improving the learning of generic forgery features.

Abstract

Generating synthetic fake faces, known as pseudo-fake faces, is an effective way to improve the generalization of DeepFake detection. Existing methods typically generate these faces by blending real or fake faces in spatial domain. While these methods have shown promise, they overlook the simulation of frequency distribution in pseudo-fake faces, limiting the learning of generic forgery traces in-depth. To address this, this paper introduces {em FreqBlender}, a new method that can generate pseudo-fake faces by blending frequency knowledge. Concretely, we investigate the major frequency components and propose a Frequency Parsing Network to adaptively partition frequency components related to forgery traces. Then we blend this frequency knowledge from fake faces into real faces to generate pseudo-fake faces. Since there is no ground truth for frequency components, we describe a dedicated training strategy by leveraging the inner correlations among different frequency knowledge to instruct the learning process. Experimental results demonstrate the effectiveness of our method in enhancing DeepFake detection, making it a potential plug-and-play strategy for other methods.

Key findings

FreqBlender significantly improves deepfake detection performance across multiple datasets, outperforming state-of-the-art methods. It effectively complements existing spatial blending methods, enhancing generalization and robustness. The method shows improved performance even on low-quality videos and against various types of deepfakes.

Approach

FreqBlender generates pseudo-fake faces by blending frequency components from real and fake faces. A Frequency Parsing Network adaptively partitions frequency components related to forgery traces, and a dedicated training strategy leverages inner correlations among different frequency knowledge.

Datasets

FaceForensics++ (FF++), Celeb-DF (CDF), DeepFake Detection Challenge (DFDC), DeepFake Detection Challenge Preview (DFDCP), and FFIW-10k (FFIW)

Model(s)

EfficientNet-b4 (for DeepFake detection), MobileNet (for face recognition), ResNet-34 (for DeepFake detection)

Author countries

China

← Previous