Capture Artifacts via Progressive Disentangling and Purifying Blended Identities for Deepfake Detection

Authors: Weijie Zhou, Xiaoqing Luo, Zhancheng Zhang, Jiachen He, Xiaojun Wu

Published: 2024-10-14 08:04:37+00:00

Comment: TCSVT(Under Review)

AI Summary

The paper introduces a novel Deepfake detection method using progressive disentangling and purifying blended identities to capture artifact features more accurately. It combines coarse- and fine-grained disentanglement strategies with an Identity-Artifact Correlation Compression (IACC) module and a separation contrastive loss. This approach allows the classifier to focus on pure artifact features, thereby enhancing detection generalization.

Abstract

The Deepfake technology has raised serious concerns regarding privacy breaches and trust issues. To tackle these challenges, Deepfake detection technology has emerged. Current methods over-rely on the global feature space, which contains redundant information independent of the artifacts. As a result, existing Deepfake detection techniques suffer performance degradation when encountering unknown datasets. To reduce information redundancy, the current methods use disentanglement techniques to roughly separate the fake faces into artifacts and content information. However, these methods lack a solid disentanglement foundation and cannot guarantee the reliability of their disentangling process. To address these issues, a Deepfake detection method based on progressive disentangling and purifying blended identities is innovatively proposed in this paper. Based on the artifact generation mechanism, the coarse- and fine-grained strategies are combined to ensure the reliability of the disentanglement method. Our method aims to more accurately capture and separate artifact features in fake faces. Specifically, we first perform the coarse-grained disentangling on fake faces to obtain a pair of blended identities that require no additional annotation to distinguish between source face and target face. Then, the artifact features from each identity are separated to achieve fine-grained disentanglement. To obtain pure identity information and artifacts, an Identity-Artifact Correlation Compression module (IACC) is designed based on the information bottleneck theory, effectively reducing the potential correlation between identity information and artifacts. Additionally, an Identity-Artifact Separation Contrast Loss is designed to enhance the independence of artifact features post-disentangling. Finally, the classifier only focuses on pure artifact features to achieve a generalized Deepfake detector.

Key findings

The proposed method demonstrates superior detection performance and better generalization across multiple benchmark datasets compared to existing state-of-the-art techniques. Ablation studies confirmed the effectiveness of each proposed component, including progressive disentanglement, the IACC module, and the Identity-Artifact Separation Contrastive loss, in enhancing model accuracy and generalization. Visualizations showed significantly improved separation of pure identity and artifact features, leading to more robust decision boundaries.

Approach

The method first performs coarse-grained disentanglement on fake faces to obtain a pair of blended identities, then fine-grainedly separates artifact features from each identity. An Identity-Artifact Correlation Compression (IACC) module, guided by information bottleneck theory, purifies identity and artifact information by reducing their potential correlations. An Identity-Artifact Separation Contrast Loss further enhances the independence of artifact features, enabling the classifier to focus solely on pure artifacts for generalized Deepfake detection.

Datasets

FaceForensics++ (FF++), Celeb-DF (CDF-v1, CDF-v2), DeepfakeDetection (DFD), Deepfake Detection Challenge (DFDC), DFDC Preview (DFDCP)

Model(s)

UNKNOWN

Author countries

China

← Previous