In Ictu Oculi: Exposing AI Created Fake Videos by Detecting Eye Blinking

Abstract

The new developments in deep generative networks have significantly improve the quality and efficiency in generating realistically-looking fake face videos. In this work, we describe a new method to expose fake face videos generated with neural networks. Our method is based on detection of eye blinking in the videos, which is a physiological signal that is not well presented in the synthesized fake videos. Our method is tested over benchmarks of eye-blinking detection datasets and also show promising performance on detecting videos generated with DeepFake.

Detecting and Simulating Artifacts in GAN Fake Images

“…In response, the forensics community has been working on detecting such generated content [4], [12], [13], [14]. Marra et al [4] propose to use raw pixel and conventional forensics features extracted from real and fake images to train a classifier.…”

Capsule-forensics: Using Capsule Networks to Detect Forged Images and Videos

“…Other methods used networks proposed by their authors [53,26,54,22,18,24,21] while others are based on a hybrid approach [55,56,57,28]. Beside deep-learning and non-deep-learning categorization, these methods could be divided into image-based classifiers [53,55,26,54,22,23,56,57,28] and video-based classifiers [18,24,20,21]. For detecting images generated by GANs, Marra et al [58] performed benchmark testing on several CNNs and proposed a statistical model for detection [59].…”

“…Automatic feature extraction has dramatically improved detection performance [14,15] while deep generative methods like generative adversarial networks (GANs) enable images [3] and videos [16,17] to be produced that are almost humanly impossible to detect as fake. The attention of the forensics community has thus shifted to these new kinds of attacks [18,19,20,21]. Several approaches are image-based [22,23] while others work only on videos frames [18,20,21] or on video frames and voice information [24].…”

“…The attention of the forensics community has thus shifted to these new kinds of attacks [18,19,20,21]. Several approaches are image-based [22,23] while others work only on videos frames [18,20,21] or on video frames and voice information [24]. Although some video-based approaches perform better than image-based ones, they are only applicable to particular kinds of attacks.…”

“…Although some video-based approaches perform better than image-based ones, they are only applicable to particular kinds of attacks. For example, many of them [18,20] may fail if the quality of the eye area is sufficiently good or the synchronization between the video and audio parts is sufficiently natural [19]. Among the image-based approaches, some are general-purpose detectors, for instance, Fridrich and Kodovsky's one [25] (applicable to both steganalysis and facial reenactment video detection) and Rahmouni et al's one [26] (applicable initially to computer-generated images and later to computer-manipulated images).…”