Feature-Integrated Deep Learning Architectures for Improved Copy-Move Forgery Recognition

Abstract

Digital image forgery has emerged as a significant concern in the era of widespread digital media, with copy-move forgery being one of the most prevalent manipulation techniques. This research investigates feature-integrated deep learning architectures for enhanced copy-move forgery detection. The primary objective is to develop and evaluate hybrid deep learning models that integrate multiple feature extraction mechanisms for improved detection accuracy. The methodology employs a quantitative experimental design utilizing benchmark datasets including CASIA, CoMoFoD, and COVERAGE, with convolutional neural networks integrated with attention mechanisms and feature fusion techniques. The hypothesis posits that feature-integrated architectures significantly outperform single-stream networks in detecting sophisticated copy-move manipulations. Results demonstrate that the proposed integrated approach achieves detection accuracy exceeding 97% on standard benchmarks, with significant improvements in detecting post-processed forgeries including rotation, scaling, and compression. The discussion reveals that multi-scale feature integration combined with attention mechanisms substantially enhances localization precision. The conclusion establishes that feature-integrated deep learning architectures represent a robust solution for copy-move forgery detection, offering superior generalization capabilities across diverse manipulation scenarios and dataset conditions.