International Journal of Science, Engineering and TechnologyAuthors: Neha Akter Alisha, Ayush Mandawgade, Divyansh Verma, Rasure Ganesh Shivkumar, Professor Rashmi Pal
Abstract: The growing ease of creating manipulated digital media, from AI-generated deepfakes to forged documents, presents a serious challenge to information integrity. This paper details a hybrid system built to counter this threat by detecting both types of forgery. Our approach combines a deep learning model, EfficientNet-B7, for spotting the subtle traces of AI generation in images, with a forensic method, Error Level Analysis (ELA), for finding manual edits in documents and PDFs. We integrated this dual-analysis engine into an interactive web application designed for straightforward use. The system delivers a clear verdict, quantitative scores, and simple visual feedback via a full- image color overlay—green for authentic media, red for suspicious content. This work demonstrates that combining distinct detection methods in a single, accessible tool offers a practical and effective solution for verifying digital media authenticity.
DOI: https://doi.org/10.5281/zenodo.17077718
