ImageGuard

The aim of the ImageGuard research project is to develop a system for the automated recognition of AI-generated and manipulated images. Since generative AI models such as Stable Diffusion, DALL-E or Midjourney are capable of image synthesis and more and more people have access to them, it is becoming increasingly difficult to distinguish real from synthetic image content and thus to keep information forgery-proof.

ImageGuard is based on a multimodal approach in which several analysis methods are combined to form a confidence value. This involves methods from classic image processing, such as the analysis of histograms or error level analysis, as well as the latest deep learning methods such as specially trained CNNs and CLIP-based feature extraction. In this way, the forgery protection of image content can be checked and evaluated in a variety of ways. How the individual metrics are to be evaluated can be found on the project website at https://www.hs-furtwangen.de/en/research/research-institutes/idacus.

The tool is implemented in a web-based application that provides a user-friendly interface for analysing images. The results of various analyses are converted into an overall confidence value using an ensemble method. Further analyses are planned, as well as the automated creation of detailed reports documenting the specific type of forgery and affected area of the image.

All in all, ImageGuard addresses the growing societal challenge of AI-generated images that are becoming increasingly sophisticated. In the future, the tool could provide journalists, consumers and educational institutions with a research basis for verifying images and help to preserve the integrity of information in the internet age.