DETECTING DIGITAL IMAGE MANIPULATION

Abstract

Digital image manipulation has become widespread, posing significant threats to journalism,forensics, law enforcement, and public trust through advanced editing techniques that evade traditional detection methods. While conventional forensic tools struggle with sophisticated tampering like splicing, copy-move, and AI-generated fakes, deep learning offers a powerful solution for accurate identification and localization. We propose Xpose, a deep learning-based stem leveraging a convolutational Neural Network (CNN) built with TensorFlow, Keras, and OpenCV in Python to detect manipulations across diverse datasets such as CASIA and COVERAGE. Comprehensive preprocessing-including normalization, resizing, data augmentation, and patch extraction-along with dropout regularization ensures robust performance, evaluated via precision, recall, F1-score, ROC curves, and confusion matrices. Xpose delivers intuitive visual outputs like heatmaps and supports scalable deployment via web interfaces with client-server integration, enhancing digital authenticity verification in real-world applications.