News coverage gains credibility if accompanied by original footage, which is widely available since many observers at the scene of events are nowadays armed with smart phones or digital cameras. Often available within a matter of seconds, such User-Generated Videos (UGVs) are a valuable resource for news services, but can also damage the reputation of a news service if it later turns out that the video had been manipulated. Therefore the authenticity of such videos needs to be verified, often under considerable time pressure. This verification process consumes significant resources in the news media industry.
Online Platform for Story Detection
In Video Veritas (InVid) is a major research initiative funded by the Horizon 2020 Program of the European Commission. The project will offer a Web-based application to automatically identify UGVs on the basis of their relevance, and confirm or reject their credibility using state-of-the-art analytical techniques.
webLyzard will lead the development of the application and the overall system integration, leveraging the existing video retrieval capabilities of its Web Intelligence platform. The goal is to identify emerging stories on social networks, automatically verify their content, and clear the professional usage rights with the content creators. This will reduce the considerable effort required to investigate the authenticity of online content, and help avoid the unintentional use of manipulated videos by news media outlets.
Certificate of Authenticity
Manual human assessment of UGVs often relies on anecdotal evidence and intuition, rather than objectively quantifiable criteria. Automating this process requires the specification of such criteria, and significant advances in knowledge processing and visual analytics. One needs to verify (i) the context of the recording – e.g., the time and location of the event, (ii) the correctness of embedded metadata – e.g., the resolution of the used camera, and (iii) the authenticity of the file – e.g., common artifacts from video editing software.
It is also important to identify the individual who recorded the video, whether this person can be contacted for verification and usage rights purposes, whether similar images of the same event are available online, and the extent to which these images are already being discussed via social media platforms.
In a first step, emerging topics will be identified by means of cluster analysis. The content of videos that are relevant in conjunction with these topics will be analyzed, annotated and indexed. Based on the annotated metadata elements (user, location, time of recording, etc.) an initial ranking will be compiled. The automatic verification process will then check whether the video was already published in a different context, and whether there are any traces of image manipulations. For videos that successfully pass this verification process, other InVid components clarify legal issues regarding the use of this video – the identification of the producer, the contractual regulation of associated rights, and a suitable remuneration agreement.
InVID will automate this workflow to provide an authenticity label for UGVs, using the data services and visual analytics tools of the webLyzard Web Intelligence Platform. Fully annotated video content will help broadcasters, news agencies, magazine editors, and bloggers to assess the trustworthiness of the material.
InVID is a project funded within the EU Horizon 2020 Programme under the project number ICT-2015-687786, carried out from 1 January 2016 to 31 December 2017. In addition to the Austrian R&D partners webLyzard and MODUL Technology, the consortium includes CERTH, the Centre for Research and Technology Hellas (Greece), the Universidad de Lleida (Spain), eXo maKina (France), and Condat (Germany). Real-time information services for journalists in press agencies and broadcasting studios will constitute the core focus of InVID. For this reason, it is particularly valuable to have three leading news organizations on board as well: Agence France Presse (AFP), Austria Press Agency (APA) and Deutsche Welle (DW).