Learning Self-Consistency for Deepfake Detection

TL;DR

This paper introduces a novel deepfake detection method leveraging source feature inconsistency, employing pair-wise self-consistency learning and an inconsistency image generator, achieving state-of-the-art accuracy on multiple datasets.

Contribution

It presents a new representation learning approach and an image synthesis method to improve deepfake detection accuracy and robustness.

Findings

Achieved 98.05% AUC in in-dataset evaluation.

Achieved 92.18% AUC in cross-dataset evaluation.

Outperformed previous methods on seven datasets.

Abstract

We propose a new method to detect deepfake images using the cue of the source feature inconsistency within the forged images. It is based on the hypothesis that images' distinct source features can be preserved and extracted after going through state-of-the-art deepfake generation processes. We introduce a novel representation learning approach, called pair-wise self-consistency learning (PCL), for training ConvNets to extract these source features and detect deepfake images. It is accompanied by a new image synthesis approach, called inconsistency image generator (I2G), to provide richly annotated training data for PCL. Experimental results on seven popular datasets show that our models improve averaged AUC over the state of the art from 96.45% to 98.05% in the in-dataset evaluation and from 86.03% to 92.18% in the cross-dataset evaluation.