Supervised Anti-Forensic DNN to Detect Multiple-time Compressed JPEG Tampered Images

Multimedia forensics is a rapidly evolving field that addresses the detection and analysis of manipulations in various forms of digital media, with a particular focus on image forgery. As the ubiquity of digital content continues to rise, so does the potential for malicious activities such as image manipulation, tampering, and forgery. This paper provides an overview of key concepts and methodologies in multimedia forensics, emphasizing techniques employed in the identification and analysis of image forgery. The first part of the paper highlights the challenges of conventional forensics methods for image forgery detection thereby focusing on multiple compressed tampered images. Depending upon this phenomenon, the paper delves into the applicability of the single-stage detection models for effective localization of the tampered regions (especially spliced regions in terms of objects) in different forms of compressed tampered images. Experiments have been conducted on our own created samples of manipulated images (i.e., manipulated in the form of splicing certain salient objects including vehicles, animals, and humans in the images). Experiment results reveal that the SSD-300 with best performed backbone network (i.e., VGG-19) can well detect the presence of spliced objects in the manipulated images with mean average precision (mAP) of 83.65%.