Enhancing Forgery Detection with ENB7IRV2 and OD-Net: A Novel Approach for CMFD Using Ensemble Feature Extraction and Keypoint Matching

With advancements in multimedia and digital image processing, digital images are frequently manipulated, raising concerns in forensic authentication. Copy-move forgery detection (CMFD) is a common technique where image regions are duplicated and altered using transformations like rotation, JPEG compression, and noise scaling. Existing methods struggle with detecting forgeries in low-contrast or smooth regions due to sparse keypoints. To address this, we propose an efficient CMFD approach. First, suspected forgery regions are identified using fast adaptive binarization (FAB) and contour detection with an Enhanced DeepLabV3Formula: see text model. Next, features are extracted using an ensemble of Enhanced EfficientNetB7 and Inception-ResNet V2 (ENB7IRV2). Finally, forgery regions are localized using the orthogonal descriptor network (OD-Net) and enhanced mean discrepancy quantization (EMDQ) approach. Our method effectively detects copy-move forgeries across diverse textures and withstands multiple attacks, providing a reliable and robust solution for multimedia forensics. The suggested technique can identify copy-move forgeries in a variety of fake photos and yield consistent findings in images with different attacks, making it a valuable and trustworthy tool. The suggested methodology significantly advances multimedia forensics, furnishing a dependable and efficient technique for identifying and pinpointing copy-move forgery within digital imagery.