The name of the game: manipulation

With the growth of 24-hour connectivity, always-on news, and social media we have access to more instantaneous information than at any time in history. Unfortunately, with the news comes “fake news”. We must rage a constant battle against the disinformation, misinformation, propaganda, and lies with which we are faced each day in our timelines. New research published in the International Journal of Multimedia Intelligence and Security, addresses one aspect of this problem – the manipulated image.

Xichen Zhang, Sajjad Dadkhah, Samaneh Mahdavifar, Rongxing Lu, and Ali A. Ghorbani of the Faculty of Computer Science, Canadian Institute for Cybersecurity at the University of New Brunswick, in Fredericton, Canada, have developed a framework that uses entity matching to analyse a suspect image and determine whether it can be verified as authentic.

The team suggests that manipulated images represent fertile ground for sowing fake news and as such must be weeded out if we are to fact check sources and validate the news that reaches us from so many disparate media. It is well known that social media users share and engage more with visual content. Moreover, it is certainly true that a picture can paint a thousand words and as such the fake news that begins with a manipulated or inappropriate image can affect our perception summarily in ways that textual content might not.

Conventionally, fact-checkers can often determine the veracity of a news item by validating the creator, the source, and evaluating the content and stance. However, a fake image associated with even the most effectively camouflaged fake news would be immediately obvious to an expert. However, there are so many images to fact check that an automated algorithmic approach has to be the way forward at least in the initial validation steps.

The new framework developed by the UNB team can retrieve valuable information and knowledge related to an image. Moreover, their statistical analysis shows the framework can offer 86 per cent accuracy. This number might well be improved significantly, but offers a solid starting point from which fake images might be weeded out. Moreover, coupled with tools that allow the textual content to be analysed in parallel, it might be possible to achieve much greater accuracy in classifying any given item and its associated images very quickly.

“Our framework is practical and effective in many different application scenarios, such as image information retrieval, image caption generation, image geo-location analysis, image tagging, stance detection between image and text content, and online image fact-checking,” the team writes.

Zhang, X., Dadkhah, S., Mahdavifar, S., Lu, R. and Ghorbani, A.A. (2022) ‘An entity matching-based image topic verification framework for online fact-checking’, Int. J. Multimedia Intelligence and Security, Vol. 4, No. 1, pp.65–85.