Image Manipulation Detection Using Man Tra-Net
Vijay Rengaraj R, R Thinakar, G. Geetha · 2023
We've created ManTraNet., a unified deep neural architecture, to battle real-world picture forgeries., which commonly entail several types and coupled modifications. ManTra-Net., in contrast to numerous other current systems., which performs both localization and detection without any further preprocessing., is an end-to-end network. ManTra-Net supports images of any resolution and a wide range of common forgery techniques., such as splices., copy-moves., deletions., and additions., as well as previously unrecognised types may be a completely convolutional network. In order to distinguish reliable picture forgery traces from other types of forgeries., we tend to design a simplistic yet efficient self-supervised learning task. Additionally., we provide a Z-score feature for native anomaly detection., frame the forgery localization disadvantage as an area anomaly detection drawback., and suggest a special memory resolution for native anomaly analysis.