Multi-modal Forgery Detection of news on International Politicians and celebrities

The quick spread of fake news is a major issue in the digital age, especially when it involves influential figures like politicians and celebrities. Manipulated images and misleading text are often used to spread misinformation. Traditional forgery detection methods focus on either image-based or textbased analysis, missing deeper connections between both. Advances in multi-modal learning enable models like ViT and BERT to jointly analyse visual and textual content. However, minor manipulations in images, such as slight facial feature changes or subtle text alterations, often go undetected. To address this, a novel multi-modal forgery detection system is proposed, integrating ResNet50, ViT, and BERT. Tesseract OCR extracts text for further processing. By analyzing both visual and textual elements at the same time, the system becomes more effective at spotting even the smallest manipulation, which would otherwise be missed. A hybrid model combines ResNet50 and ViT for images, while BERT detects textual inconsistencies. The extracted features are combined and classified into real/fake and manipulation types, improving accuracy in detecting subtle forgeries.