Interpretable Face Manipulation Detection via Feature Whitening

TL;DR

This paper introduces an interpretable face manipulation detection method that enhances trustworthiness by making the detection process transparent through feature whitening, balancing accuracy and interpretability.

Contribution

It proposes a novel feature whitening module that decorrelates features to improve interpretability without sacrificing detection accuracy.

Findings

Balances detection accuracy and interpretability

Feature whitening improves model transparency

Achieves trustworthy face manipulation detection

Abstract

Why should we trust the detections of deep neural networks for manipulated faces? Understanding the reasons is important for users in improving the fairness, reliability, privacy and trust of the detection models. In this work, we propose an interpretable face manipulation detection approach to achieve the trustworthy and accurate inference. The approach could make the face manipulation detection process transparent by embedding the feature whitening module. This module aims to whiten the internal working mechanism of deep networks through feature decorrelation and feature constraint. The experimental results demonstrate that our proposed approach can strike a balance between the detection accuracy and the model interpretability.