Deceptive Beauty: Evaluating the Impact of Beauty Filters on Deepfake and Morphing Attack Detection

View on arXiv ← Back to list

Authors: Sara Concas, Simone Maurizio La Cava, Andrea Panzino, Ester Masala, Giulia Orrù, Gian Luca Marcialis

Published: 2025-09-17 15:59:44+00:00

AI Summary

This research investigates the impact of beauty filters on deepfake and morphing attack detection. The study evaluates state-of-the-art detectors' performance on benchmark datasets before and after applying various filters, revealing significant performance degradation. This highlights vulnerabilities introduced by facial enhancements and underscores the need for more robust detection models.

Abstract

Digital beautification through social media filters has become increasingly popular, raising concerns about the reliability of facial images and videos and the effectiveness of automated face analysis. This issue is particularly critical for digital manipulation detectors, systems aiming at distinguishing between genuine and manipulated data, especially in cases involving deepfakes and morphing attacks designed to deceive humans and automated facial recognition. This study examines whether beauty filters impact the performance of deepfake and morphing attack detectors. We perform a comprehensive analysis, evaluating multiple state-of-the-art detectors on benchmark datasets before and after applying various smoothing filters. Our findings reveal performance degradation, highlighting vulnerabilities introduced by facial enhancements and underscoring the need for robust detection models resilient to such alterations.

Key findings

Beauty filters caused a significant decrease in the performance of deepfake and morphing attack detectors. The impact varied depending on the filter intensity, model architecture, and type of manipulation. This demonstrates the need for filter-robust deepfake and morph detection models.

Approach

The researchers applied smoothing filters with varying intensities to benchmark datasets of deepfakes and morphs. They then evaluated the performance of pre-trained AlexNet and VGG19 models on both the original and filtered data, measuring metrics like Equal Error Rate (EER) and Area Under the ROC Curve (AUC).

Datasets

CelebDF (deepfakes) and AMSL (morphing attacks)

Model(s)

AlexNet and VGG19

Author countries

Italy

← Previous