Symmetric non negative matrices factorization applied to the detection of communities in graphs and forensic image analysis

With the proliferation of data, particularly on social networks, the accuracy of the information becomes uncertain. In this context, a major challenge lies in detecting image manipulations, where alterations are made to deceive observers. Aligning with the anomaly detection issue, recent methods approach the detection of image transformations as a community detection problem within graphs associated with the images. In this study, we propose using a community clustering method based on non-negative symmetric matrix factorization. By examining several experiments detecting alterations in manipulated images, we assess the method’s robustness and discuss potential enhancements. We also present a process for automatically generating visually and semantically coherent forged images. Additionally, we provide a web application to demonstrate this process.