Deepfake Video Detection Using Recurrent Neural Networks

Güera, David; Delp, Edward J.

doi:10.1109/avss.2018.8639163

Cited by 780 publications

(424 citation statements)

References 24 publications

Supporting

Mentioning

418

Contrasting

Unclassified

Order By: Relevance

“…Other approaches, rather than leveraging semantically-meaningful cues, use machine learning and neural networks to distinguish GAN from real images. Marra et al [10] use a network based on XceptionNet, Hsu et al [5] develop a deep forgery discriminator with a contrastive loss function, and Guera and Delp [4] use recurrent neural networks to detect GAN video. A key concern with methods based on deep networks is that they could easily be incorporated into the GAN's discriminator and, with additional training, the generator could be fine-tuned in order to learn a countermeasure for any differentiable forensic.…”

Section: Related Workmentioning

confidence: 99%

Detecting GAN-Generated Imagery Using Saturation Cues

McCloskey

Albright

2019

2019 IEEE International Conference on Image Processing (ICIP)

172

View full text Add to dashboard Cite

Image forensics is an increasingly relevant problem, as it can potentially address online disinformation campaigns and mitigate problematic aspects of social media. Of particular interest, given its recent successes, is the detection of imagery produced by Generative Adversarial Networks (GANs), e.g. 'deepfakes'. Leveraging large training sets and extensive computing resources, recent work has shown that GANs can be trained to generate synthetic imagery which is (in some ways) indistinguishable from real imagery. We analyze the structure of the generating network of a popular GAN implementation [7], and show that the network's treatment of color is markedly different from a real camera in two ways. We further show that these two cues can be used to distinguish GAN-generated imagery from camera imagery, demonstrating effective discrimination between GAN imagery and real camera images used to train the GAN.

show abstract

Section: Related Workmentioning

confidence: 99%

Detecting GAN-Generated Imagery Using Saturation Cues

McCloskey

Albright

2019

2019 IEEE International Conference on Image Processing (ICIP)

172

View full text Add to dashboard Cite

show abstract

“…At this point, it should be mentioned that recently, AI-synthesized face swapping videos-commonly known as the DeepFakes-have become an emerging problem. A machine learning based free software tool has made it easy to create believable face swaps in videos, leaving few traces of manipulation [22], and correspondingly, there is an increasing interest in developing algorithms that can detect them. However, it should be noted that in many LBS scenarios and applications, what is important is ensuring that a (and not 'the') human user holds a device that is close to an area of interest, while the identity of the user per se is not of much importance.…”

Section: A Proposal For Strong Identitiesmentioning

confidence: 99%

QuietPlace: An Ultrasound-Based Proof of Location Protocol with Strong Identities

Kounas

Voutyras

Palaiokrassas

et al. 2020

ASI

View full text Add to dashboard Cite

Location-based services are becoming extremely popular due to the widespread use of smartphones and other mobile and portable devices. These services mainly rely on the sincerity of users, who can spoof the location they report to them. For applications with higher security requirements, the user should be unable to report a location different than the real one. Proof of Location protocols provide a solution to secure localization by validating the device's location with the help of nearby nodes. We propose QuietPlace, a novel protocol that is based on ultrasound and provides strong identities, proving the location of the owner of a device, without exposing though their identity. QuietPlace provides unforgeable proof that is able to resist to various attacks while respecting the users' privacy. It can work regardless of certificate authority and location-based service and is able to support trust schemas that evaluate the participants' behavior. We implement and validate the protocol for Android devices, showing that ultrasound-based profiles offer a better performance in terms of maximum receiving distance than audible profiles, and discuss its strengths and weaknesses, making suggestions about future work.

show abstract

“…Researcher has too quickly and efficiently used a method of detecting facial manipulation in videos and focuses in particular on two new techniques used to create hyper-realistic faked images: Deepfake and Face2Face. David Guera [8] et al Proposed a system that uses a convolutionary neural network(CNN) to select characteristics at frame level. Then, these functions are used to train current neural network (RNN), which learns to determine whether or not a video is subject to manipulation.…”

Section: Literature Reviewmentioning

confidence: 99%

“…By studying various techniques in literature survey we will improve the methods technique. Using Various Techniques above mentioned we will improvise the accuracy of the fake videos identification [8,5]. We will analyze all the parameters to improve the accuracy .Our Results will be based on the accuracy of each technique and on average of all the techniques we will determine whether the tested video is real or fake.…”

Section: Fig2 Same Videos Different Traffic Flowmentioning

confidence: 99%

Detecting fake Videos using Block Chain and Smart Contracts

2020

IJRTE

View full text Add to dashboard Cite

Abstract: The Rapid growth of agile gadgets has led to tremendous increase in digital media utilization, mostly for mobile video in ease of marketing. As encryption provides better user confidentiality and perseveration, greater number of online movement is associated with end-to-end encryption form. Irrevlant content such as unreal, violent and unconstitutional videos are being circulated online without being identified its truthfulness creating a platform for intruders and attackers. It is necessary for users to identify and report the contents of the video. Sometimes these videos act as evidence in courts to prove the guilty and the proper state and contents recorded in it .We aim a system to detect and classify the video's truthfulness to solve these problems. A detailed collection of studies has been toted to demonstrate the efficacy of new program over current literature.

show abstract

Deepfake Video Detection Using Recurrent Neural Networks

Cited by 780 publications

References 24 publications

Detecting GAN-Generated Imagery Using Saturation Cues

Detecting GAN-Generated Imagery Using Saturation Cues

QuietPlace: An Ultrasound-Based Proof of Location Protocol with Strong Identities

Detecting fake Videos using Block Chain and Smart Contracts

Contact Info

Product

Resources

About