As digital image manipulation becomes increasingly sophisticated, ensuring image authenticity has become a critical requirement, particularly for embedded and Internet of Things (IoT) systems that demand real-time and reliable decision-making. This work proposes a hybrid deep learning framework that performs image forgery detection and localization through a clear, stepwise process. First, EfficientNet-B0 is employed for high-accuracy detection of forged images. Second, a U-Net architecture enhanced with a spatial attention mechanism is used to localize manipulated regions at the pixel level. Third, the complete model is optimized for IoT deployment through compression and quantization to reduce computational load while preserving accuracy. The framework is rigorously validated on three benchmark datasets - CASIA 1.0, CASIA 2.0, and CoMoFoD v2. Across these datasets, the method achieves exceptional performance, including 99.84% accuracy for copy-move detection on CASIA 1.0, 99.81% for splicing detection on CASIA 1.0, 99.97% accuracy on CASIA 2.0, and 99.96% accuracy on CoMoFoD v2, alongside precise region localization. Finally, comparative analysis confirms that the proposed approach consistently surpasses state-of-the-art methods in both detection reliability and localization accuracy. By integrating high-performing models with IoT-friendly optimizations, this work provides a robust and scalable deployment-oriented solution for digital image authenticity verification, with experiments conducted in a simulated edge setting and on-device benchmarking identified as future work.
Sharma et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: