Fighting misinformation: Image forgery detection in social media streams (Master Thesis, Finished)
Information has never been so abundant and so easily accessible as it is nowadays. Yet, many authors and observers argue that there have never been so many misinformed, which has led some people to dub our time as "The Age of Misinformation". In the past few years, the issue of fake news, that is, disinformation disseminated via social media channels and the Internet with the intent to alter public perception of a particular topic, has gained widespread attention, especially in the context of politics.
Traditionally, disinformation in news reporting is countered by fact-checking, as performed by many existing fact-checking organisations and similar initiatives. However, this mode of exposing falsified news involves manual labor by journalists and volunteers and is therefore prone to error and bias and unable to cope with the sheer volume of information generated by the different actors. A potential solution could involve automatic detection of misinformation in news articles using techniques from multimedia analysis and machine learning.
In this vein, this Master's Thesis aims at identifying and correctly classifying forged images (e.g. photomontages, retouched images) as used in news articles and social media posts. The concrete use-case will involve a Twitter stream and one focus of the Thesis will lie on the "real-time" aspect. Furthermore, we will experiment with and compare different techniques for image forgery detection and evaluate them in terms of accuracy and computational requirements.
Start / End Dates
2019/01/02 - 2019/07/01