A Photo Forensics-Based Prototype to Combat Revenge Porn

摘要

The proliferation of image-and video-sharing websites has increased revenge porn images. In this paper, we propose a photo forensic-based prototype to aid revenge porn victims to automatically retrieve offending images. Our prototype combines a deep learning-based nude detection filter with a face recognition filter to find if a nude image containing the victim's face is posted on targeted social media accounts. After finding revenge porn images from the targeted accounts, our prototype iteratively finds other accounts containing potential revenge porn images. Revenge porn images from these newly identified accounts are then filtered out. Experiment with 1254 images shows that our method can filter revenge porn images in 91% cases.