The falsification of faces in videos is a growing phenomenon over the years. One of the most popular ways to tamper a face in a video is known as “deepfake”, Today, many tools exist to allow anyone to create a deepfake to discredit an individual or usurp an identity. Fortunately, the detection of deepfakes is an increasing topic of interest for the scientific community. As a result, many efforts have been made to develop mechanisms to automatically identify deepfake videos. In addition, several public deepfakes datasets have been built to help researchers to develop more effective detection methods. The most recent and also the most complete of these datasets is the one built by Facebook as part of the international DeepFake Detection Challenge (DFDC). Thousands of different frameworks, mainly based on deep learning, have been proposed during this challenge. The best solution that has been proposed obtains the accuracy of 82% on the DFDC dataset. However, the accuracy of this method is only 65% on unseen videos from the Internet. In this paper we analyse the five best methods of the DFDC and their complementarity. In addition, we experimented different assembly strategies (boosting, bagging and stacking) among these solutions. We show that we can achieve a large improvement $(+ 41\%$ on log loss and $+2.26\%$ on accuracy) when we carefully choose the models to be assembled with the most appropriate right merging method to use.

The progress made in deep learning has allowed the deployment of more powerful biometric authentication systems instead of traditional ones based on passwords or PIN codes. Facial recognition is widely used on smartphones to grant user access. However, advances in deep learning also improve methods for doctoring images and videos. A fraudulent user can use these methods to steal the identity of another person. It is very easy for impostors to present to the smartphone an image or video of the victim's face displayed on another screen. In this paper, we describe the security risks when a facial recognition system is attacked by presenting an image, a video or an interactive deepfake displayed on a screen. We also present a deep learning-based method to detect this kind of attack.

With the 2019 Coronavirus pandemic, we have seen an increasing use of remote technologies such has remote identity verification.The authentication of the user identity is often performed through a biometric matching of a selfie and a video of an official identity document.In such a scenario, it is essential to verify the integrity of both the selfie and the video.In this article, we propose a method to detect double video compression in order to verify the video integrity.We will focus on the H.264 compression which is one of the mandatory video codecs in the WebRTC Requests For Comments.H.264 uses an integer approximation of the Discrete Cosine Transform (DCT).Our method focuses on the DCT coefficients to detect a double compression.The coefficients roughly follow a Laplacian distribution, we will show that the distribution parameters vary with respect to the quantisation parameter used to compress the video.We thus propose a statistical hypothesis test to determine whether or not a video has been compressed twice.

AbstractIn this chapter, we present the various categories of Face Manipulation and their use within different remote operational systems. We then use the example of remote identity document onboarding systems to illustrate how each category can be used in practice to compromise such a system. After a definition of the different Face Manipulation categories and the common algorithms used to produce them, we go through the various manipulation detection algorithms and common image and video forgery datasets. We then introduce some known counter-forensics methods that can be used by an attacker to avoid detection. Knowing the detection methods and the counter-forensic, we present how we can build up a safer system by using the correct methods at the correct time. But also how knowledge about the tampering process could be used to design the user experience to make the systems harder to compromise. We complete this review by the standardisation effort and legal aspect on the matter. And we conclude by discussing the remaining challenges and perspectives for better use of nowadays detection methods in practical usage.

Copy-Move forgery has been widely studied as it is a really common forgery. Furthermore, it is the easiest forgery to create with serious security-related threats in particular for distant remote id onboarding where company ask their customer to send a photo of their ID document. It is then easy for a counterfeit to alter the information on the document by copying and pasting letters within the photo. On the other hand, copy-move detection algorithms are known to perform worse in presence of similar but genuine objects preventing us from using them in practical situations like remote ID on boarding. In this article we propose a novel copy-move public dataset containing forged ID documents and study current state-of-the-art performances on this dataset to evaluate their potential use in practical situations.

While creating digital image forgeries, it is often necessary to hide an object from the image. For instance, to replace a person on a picture one would first remove the person on the image before inserting the new one. This process of suppression can be performed in many ways but almost always involve recreating some background textures. While recreating it, artists sometimes use common tools and apply smooth transitions to seamlessly blend the result. This operation can decrease the created texture sharpness. In this paper, we explore the possibility to reveal this effect to expose possible forgeries.

It is becoming harder and harder to believe in digital images authenticity. This is due to the increasing ease of producing forged images using retouching software more powerful than ever. On software like Photoshop, one can easily remove an element from an image with just a few clicks. Copy-move is one of the elemental forgery methods that consists in the duplication of one portion of the image. Such manipulation is very usual and dangerous in official document falsification (ID, breeder docs, bills. . . ). In this paper, we propose a novel method for the detection of Copy and Move forgery while maintaining a low false positive rate.

This paper presents a novel dataset for image and face manipulation detection and localization called DEFACTO. The dataset was automatically generated using Microsoft common object in context database (MSCOCO) to produce semantically meaningful forgeries. Four categories of forgeries have been generated. Splicing forgeries which consist of inserting an external element into an image, copy-move forgeries where an element within an image is duplicated, object removal forgeries where objects are removed from images and lastly morphing where two images are warped and blended together. Over 200000 images have been generated and each image is accompanied by several annotations allowing precise localization of the forgery and information about the tampering process.

Onboarding of new customers is a sensitive task for various services, like Banks who have to follow the Know Your Customer (KYC) rules. Mobile Onboarding Applications or KYC by Streaming are expanding rapidly to provide this capacity at home. Unfortunately, this leaves the authentication tools in the hand of end-users, allowing the attacker to directly tamper the video stream. With the rise of new digital face manipulation technologies, traditional face spoofing attacks such as presentation attacks or replay attacks should not be the only one to be considered. A new kind of face spoofing attacks (i.e. digital face spoofing) needs to be studied carefully. In this paper, we analyze those new kinds of attacks and propose a method to secure identity documents against both the traditional attacks and the new ones.

The QR (Quick Response) code is a two-dimensional barcode, which was designed for storage information and highspeed reading applications. Being cheap to produce and fast to read, it becomes actually a popular solution for product labeling. Ones try to make QR code a solution against counterfeiting. In this paper, we present a novel technique that permits to create a secure printed QR code which is robust against Scan & Reprint attack. The code is constructed by replacing the background of the standard one by a specific textured pattern which does not affect the normal reading of the encoded message. Scan & Reprint attacks lead to the degradation of the texture and change its statistical characteristics, which can be detected thanks to a statistical hypothesis testing. The experimental results show a good performance of the proposed solution.

Picture falsification on identity documents is a recurring problem. Text falsification can be mitigated on printed documents thanks to digital signature, but for picture the only safe strategy was to integrate an expensive electronic chip in the document. This paper proposes a low-cost alternative, allowing to check offline the authenticity of the image thanks to digitally signed characteristics extracted from the picture.

Distribution of smartphone apps represents a major risk for tomorrow. Alternative app stores filled by pirated content create a dangerous substrate for the diffusion of tricky apps. We present in this paper a dedicated Crawling Engine conceived to protect distribution of genuine apps.

DOCSCOPE is an ANR project dedicated to the exploration of new approaches to authenticate ID documents and to detect falsification or counterfeiting. In this paper we focus on the usage of printing techniques signature.

The european ESPRIT project STRETCH is dedicated to the management of documents by their content. We present in this paper the high-performance developments and the programming support required by STRETCH indexing and retrieval processes.

In this paper we introduce T++: a parallel language with object-oriented features designed for Multi-SIMD parallel computers. We propose a new approach to express simultaneously task and data parallelism. We describe the advantages of an object-oriented approach and what kind of semantics we choose to structure our task-data-parallelism. Finally, we explain how to implement it efficiently on a proprietary Multi-SIMD architecture: the SYMPHONIE concept.