Executive Summary : | The rise of technology has made it easier for people to access and share images and videos on social media, but there is a growing concern about the potential leakage of sensitive documents, such as exam papers, trade secrets, and copyrighted materials. To address this issue, researchers are exploring source camera identification (SCI) techniques for captured document images, which can help identify potential leakage sources. Deep learning approaches have shown significant progress in various domains, and they can be used to develop effective SCI methods for document images in real-world scenarios. However, current techniques are limited in their exploration of source camera model identification for smartphone cameras and natural scene images, with little consideration for social media sharing aspects. To develop effective SCI methods, researchers propose using appropriate patch selection and pre-processing strategies, multi-branch networks, and large datasets of document images, particularly examination papers. This project will collect and process these images using different mobile devices and social media platforms, conducting detailed experiments and exploring comparative analysis approaches for better efficacy. The aim is to deepen the understanding of explainability analysis and incremental learning in document image forensics for potential adoption in real-world situations. |