Publication

Abstract : The growing prevalence of digitally manipulated content calls for effective image forgery detection methods. This study proposes a hybrid approach that integrates Error Level Analysis (ELA) with the DeepLabV3 segmentation model. ELA identifies compression artifacts caused by tampering, providing visual cues for forgery detection. These ELA maps are fused with the original images to form a multi-channel input, enabling DeepLabV3 to capture detailed forgery patterns. The system is evaluated on the CASIA v2 dataset using ground truth masks for supervised training, ensuring precise detection and localization of tampered regions. Metrics such as Intersection over Union (IoU), precision, recall, and F1-score validate the model’s effectiveness. The results highlight the potential of combining traditional forensic techniques with advanced deep learning for reliable and scalable image forgery detection in real-world scenarios.