CONFERENCE / ICCAIS-2026

The rapid evolution of image editing software and the extensive use of social media have led to a significant increase in digitally manipulated images, raising serious concerns regarding the reliability and authenticity of visual information. Conventional image forgery detection methods are typically designed to identify specific manipulation patterns and often fail to maintain robustness under real-world post-processing operations. Although Convolutional Neural Networks (CNNs) have achieved notable success in image analysis, accurately identifying subtle and well-blended forgeries remains a challenging task. This paper proposes a hybrid digital image forgery detection framework that combines Error Level Analysis (ELA) with a custom-designed CNN architecture. ELA is utilized as a preprocessing mechanism to expose compression inconsistencies introduced during image tampering, while the CNN learns discriminative forensic features from the resulting ELA representations. By integrating compression-based cues with deep feature learning, the proposed approach enhances the detection of both conspicuous and imperceptible manipulations. The effectiveness of the method is validated on the CASIA2 image dataset, where it achieves a detection accuracy of 96.7%, outperforming a baseline transfer learning model. Experimental findings confirm that the inclusion of ELA improves the robustness and generalization capability of CNN-based forgery detection systems.