Transformation on Computer-Generated Facial Image to Avoid Detection by Spoofing Detector

TL;DR

This paper introduces a CNN-based method that transforms CG facial images to evade detection by spoofing detectors, highlighting vulnerabilities in facial authentication systems.

Contribution

It proposes a novel autoencoder-transformer CNN approach trained with a black-box discriminator to increase the naturalness of CG images for spoofing evasion.

Findings

Over 50% of transformed images bypassed detectors

Highlights security risks in facial authentication systems

Demonstrates effectiveness of the proposed transformation method

Abstract

Making computer-generated (CG) images more difficult to detect is an interesting problem in computer graphics and security. While most approaches focus on the image rendering phase, this paper presents a method based on increasing the naturalness of CG facial images from the perspective of spoofing detectors. The proposed method is implemented using a convolutional neural network (CNN) comprising two autoencoders and a transformer and is trained using a black-box discriminator without gradient information. Over 50% of the transformed CG images were not detected by three state-of-the-art spoofing detectors. This capability raises an alarm regarding the reliability of facial authentication systems, which are becoming widely used in daily life.