Security of Text to Image Conversions

Abstract

The use of images and icons to represent news or narratives has grown in popularity. Still, one critical problem is that they are not equivalent to language, making them vulnerable to adversary attacks. This study examines the impact of image-poisoning attacks based on polysemantic words and of image attacks based on cultural differences when converting text to images. Such attacks can lead to the loss of important information and create confusion and incorrect interpretations of the intended meaning, misinforming the general public. The study specifically focuses on possible effects in a news and story context. This study highlights the significance of taking security considerations into account when image-based attacks are relevant and motivates the development of strategies to ensure that information is conveyed through images and icons in a culturally appropriate and accurate manner, as well as to prevent image tampering and the spread of false information by attackers.