Deepfake Face Traceability with Disentangling Reversing Network

View on arXiv ← Back to list

Authors: Jiaxin Ai, Zhongyuan Wang, Baojin Huang, Zhen Han

Published: 2022-07-08 03:05:28+00:00

AI Summary

This paper introduces a novel deepfake face traceability approach, addressing the limitation of current deepfake detection methods that only identify fakes without tracing their origin. The authors propose a disentangling reversing network that infers the original face from a deepfake by decoupling latent space features.

Abstract

Deepfake face not only violates the privacy of personal identity, but also confuses the public and causes huge social harm. The current deepfake detection only stays at the level of distinguishing true and false, and cannot trace the original genuine face corresponding to the fake face, that is, it does not have the ability to trace the source of evidence. The deepfake countermeasure technology for judicial forensics urgently calls for deepfake traceability. This paper pioneers an interesting question about face deepfake, active forensics that know it and how it happened. Given that deepfake faces do not completely discard the features of original faces, especially facial expressions and poses, we argue that original faces can be approximately speculated from their deepfake counterparts. Correspondingly, we design a disentangling reversing network that decouples latent space features of deepfake faces under the supervision of fake-original face pair samples to infer original faces in reverse.

Key findings

The proposed network successfully reverses original faces from deepfakes, with qualitative and quantitative results demonstrating high fidelity between original and traced faces. However, performance is impacted by factors such as large differences in skin color or gender between original and fake faces, and low-quality deepfakes.

Approach

The proposed disentangling reversing network consists of an identity disentangling module and a face reversing module. The former disentangles identity features from other attributes, while the latter maps fake facial identities to original ones, generating a traced face. The network is trained using a combination of loss functions including content loss, contrast loss, and cycle consistency loss.

Datasets

Celeb-DF-v2 and FaceForensics++

Model(s)

A disentangling reversing network based on UNet architecture, using a pretrained ResNet-50 face recognition network for supervision.

Author countries

UNKNOWN

← Previous