GAN-generated Faces Detection: A Survey and New Perspectives

Abstract

Generative Adversarial Networks (GAN) have led to the generation of very realistic face images, which have been used in fake social media accounts and other disinformation matters that can generate profound impacts. Therefore, the corresponding GAN-face detection techniques are under active development that can examine and expose such fake faces. In this work, we aim to provide a comprehensive review of recent progress in GAN-face detection. We focus on methods that can detect face images that are generated or synthesized from GAN models. We classify the existing detection works into four categories: (1) deep learning-based, (2) physical-based, (3) physiological-based methods, and (4) evaluation and comparison against human visual performance. For each category, we summarize the key ideas and connect them with method implementations. We also discuss open problems and suggest future research directions.

The Face Deepfake Detection Challenge

“…The task of detecting images with human faces created by Generative Adversarial Networks (GAN) has attracted significant attention from the research community, as witnessed by some recent surveys on the topic [ 4 , 5 , 6 ].…”

“…After the seminal work by Do et al [ 13 ], which was entirely data-driven, other approaches followed where images are pre-processed (e.g., by high-pass filtering [ 14 ], or working on the chrominance components [ 15 ]) in order to let the network work on a facilitating domain. Several approaches were then proposed [ 4 ], and it seemed that, as suggested by Wang et al, most GAN-generated images shared common flaws that made them easy to detect [ 16 ]. However, it must be that such flaws were progressively reduced, given that a recent study by Gagnaniello et al [ 17 ] shows that GAN-detection methods are apparently still far from showing reliable performance, especially when tested images that differ significantly from those in the training set.…”

Fusing Global and Local Features for Generalized AI-Synthesized Image Detection

“…Notable examples of AI-synthesized media include the high-quality images synthesized by the Generative Adversarial Networks (GANs) [1], and videos with face swaps or puppetry created by the auto-encoder models [2]. The synthetic images/videos have become challenging for humans to distinguish [3], and correspondingly, a slew of detection methods [4,5] have been developed to mitigate the potential risks posed by such AI-synthesized images.…”

Open-Eye: An Open Platform to Study Human Performance on Identifying AI-Synthesized Faces

“…As the existing methods are either less efficient or not accurate enough to handle the torrent of daily uploads of the public content [30], the users must have the ability to rec-ognize the fake faces from the real ones. Recently, the studies investigating the human performance of AI-synthesized faces detection have been conducted [19,26].…”