Dual-Module Architecture for Robust Image Forgery Segmentation and Classification Toward Cyber Fraud Investigation

This study presents a dual-module architecture for image forgery detection in the context of cyber fraud investigation, designed to provide interpretable and court-admissible forensic evidence. The forgery segmentation module built on an encoder–decoder structure segments forged regions at the pixel level to produce a binary mask. The forgery classification module with two-stream structure integrates contextual and noise-residual cues from the raw image and the binary mask to determine the designated forgery method. The segmentation module achieves an F1-Score of 0.875 and an IoU of 0.78, while the classification module reaches an F1-Score of 0.94. The combined system attains an end-to-end F1-Score of 0.855 and AUC of 0.91, demonstrating reliable detection performance and enhanced explainability. These results highlight the framework’s potential for forensic image analysis and its practical applicability to real-world cyber fraud investigations.