This paper presents a method to automatically and efficiently detect face tampering in videos, and particularly focuses on two recent techniques used to generate hyper-realistic forged videos: Deepfake and Face2Face. Traditional image forensics techniques are usually not well suited to videos due to the compression that strongly degrades the data. Thus, this paper follows a deep learning approach and presents two networks, both with a low number of layers to focus on the mesoscopic properties of images. We evaluate those fast networks on both an existing dataset and a dataset we have constituted from online videos. The tests demonstrate a very successful detection rate with more than 98% for Deepfake and 95% for Face2Face.

https://hal-upec-upem.archives-ouvertes.fr/hal-01867298/document

MesoNet: a Compact Facial Video Forgery Detection Network

MesoNet: a Compact Facial Video Forgery Detection Network.

Validating a given multimedia content is nowadays quite a hard task because of the huge amount of possible alterations that could have been operated on it. In order to face this problem, image and video experts have proposed a wide set of solutions to reconstruct the processing history of a given multimedia signal. These strategies rely on the fact that non-reversible operations applied to a signal leave some traces ("footprints") that can be identified and classified in order to reconstruct the possible alterations that have been operated on the original source. These solutions permit also to identify which source generated a specific image or video content given some device-related peculiarities. The paper aims at providing an overview of the existing video processing techniques, considering all the possible alterations that can be operated on a single signal and also the possibility of identifying the traces that could reveal important information about its origin and use.

/pdf/an-overview-on-video-forensics-20693kkcsx.pdf

An overview on video forensics

The illegal distribution of a digital movie is a common and significant threat to the film industry. With the advent of high-speed broadband Internet access, a pirated copy of a digital video can now be easily distributed to a global audience. A possible means of limiting this type of digital theft is digital video watermarking whereby additional information, called a watermark, is embedded in the host video. This watermark can be extracted at the decoder and used to determine whether the video content is watermarked. This paper presents a review of the digital video watermarking techniques in which their applications, challenges, and important properties are discussed, and categorizes them based on the domain in which they embed the watermark. It then provides an overview of a few emerging innovative solutions using watermarks. Protecting a 3D video by watermarking is an emerging area of research. The relevant 3D video watermarking techniques in the literature are classified based on the image-based representations of a 3D video in stereoscopic, depth-image-based rendering, and multi-view video watermarking. We discuss each technique, and then present a survey of the literature. Finally, we provide a summary of this paper and propose some future research directions.

An Overview of Digital Video Watermarking

Ternary radial harmonic Fourier moments based robust stereo image zero-watermarking algorithm

As the 3D content market is growing up rapidly, copyright issues on the 3D content naturally have come into play. Among the various 3D coding techniques, depth-imagebased rendering (DIBR) has been studied by numerous researchers due to its coding efficiency and 2D compatibility. However, conventional watermarking scheme for 2D content cannot be directly applied to the DIBR content because DIBR process is considered as a new type of attack in terms of watermarking design. In this paper, we propose a watermarking scheme for DIBR 3D videos with an improved version of DT-CWT magnitude quantization. In order for efficient watermark embedding and extraction on DIBR 3D videos, all frames are grouped. Then, the watermark is embedded repeatedly and periodically so that extraction performance is enhanced. Furthermore, the quantization process is improved to a faster version by checking sum of magnitude and setting a start point of quantization order optimally. The simulation results show that the embedded watermark is stably extracted from center views and virtual views. Moreover, when the watermarked virtual views are distorted or processed by compression, scaling, median filtering and noise addition, the embedded watermark is extracted with low BERs.

https://www.actapress.com/PDFViewer.aspx?paperId=455004

Dibr 3d video watermarking with faster dt-cwt quantization

This paper proposes a composite mask based watermarking method, which improves perceptual quality of videos watermarked by traditional video watermarking scheme. The composite mask including noise visibility function (NVF) mask, adaptive dithering mask, and contour mask considers human visual system (HVS). The adaptive dithering mask solves the block artifact problem caused by expanding the watermark basic pattern for the robustness against various downscaling attacks. The contour mask makes up for the disadvantage of NVF mask, which cannot handle separately high textured regions and contour or edge regions. Extensive experiments prove that the watermarking method based on the proposed composite mask satisfies imperceptibility, robustness against downscaling as well as common video processing, and real-time performance.

https://hklee.kaist.ac.kr/publications/2010%20IPTA(with%20Oh%20Tae-Woo).pdf

Enhancing perceptual quality of watermarked high-definition video through composite mask

A point cloud is a collection of points, which is measured by time-of-flight information from LiDAR sensors, forming geometrical representations of the surrounding environment. With the algorithmic success of deep learning networks, point clouds are not only used in traditional application domains like localization or HD map construction but also in a variety of avenues including object classification, 3D object detection, or semantic segmentation. While point cloud analytics are gaining significant traction in both academia and industry, the computer architecture community has only recently begun exploring this important problem space. In this paper, we conduct a detailed, end-to-end characterization on deep learning based point cloud analytics workload, root-causing the frontend data preparation stage as a significant performance limiter. Through our findings, we discuss possible future directions to motivate continued research in this emerging application domain.

Characterization and Analysis of Deep Learning for 3D Point Cloud Analytics

Cleidocranial dysplasia (CCD) is an autosomal-dominant disease characterized by the delayed closure of cranial sutures, defects in clavicle formation, supernumerary teeth, and delayed tooth eruption. Defects in the Runt-related transcription factor 2 (RUNX2), a master regulator of bone formation, have been identified in CCD patients. The aim of this study was to identify the molecular genetic causes in a CCD family with delayed tooth eruption. The 23-year-old female proband and her mother underwent clinical and radiographic examinations, and all coding exons of the RUNX2 were sequenced. Mutational analysis revealed a single nucleotide deletion mutation (NM_001024630.4 : c.357delC) in exon 3 in the proband and her mother. The single C deletion would result in a frameshift in translation and introduce a premature stop codon [p.(Asn120Thrfs*24)]. This would result in the impaired function of RUNX2 protein, which may be the cause of delayed eruption of permanent teeth in the family.

/pdf/a-novel-runx2-mutation-in-a-korean-family-with-cleidocranial-3zxd08t27b.pdf

A Novel RUNX2 Mutation in a Korean Family with Cleidocranial Dysplasia

Due to the development of video understanding and big data analysis research field using deep learning technique, intelligent machines have replaced the tasks that people performed in the past in various fields such as traffic, surveillance, and security area. In the sports field, especially in soccer games, it is also attempting quantitative analysis of players and games through deep learning or big data analysis technique. However, because of the nature of soccer analysis, it is still difficult to make sophisticated automatic analysis due to technical limitations. In this paper, we propose a deep learning based motion recognition technique which is the basis of high level automatic soccer analysis. For sophisticated motion recognition, we maximize recognition accuracy by sequentially processing the data in three steps: data acquisition, data augmentation, and 3D CNN based motion classifier learning. As can be seen from the experimental results, the proposed method guarantees real-time speed performance and satisfactory accuracy performance.

/pdf/soccer-object-motion-recognition-based-on-3d-convolutional-2noql1gwaj.pdf

Jiwon Lee

Papers

Dibr 3d video watermarking with faster dt-cwt quantization

Enhancing perceptual quality of watermarked high-definition video through composite mask

Characterization and Analysis of Deep Learning for 3D Point Cloud Analytics

A Novel RUNX2 Mutation in a Korean Family with Cleidocranial Dysplasia

Soccer object motion recognition based on 3D convolutional neural networks