Multiscale Feature Fusion for Robust Copy-Move Forgery Detection in Digital Images

The detection of copy-move image forgeries has been getting significant attention due to the ever-increasing use of digital images in various fields, but the techniques being developed are quite unable to cope with subtle forgeries, mainly in low-resolution images. In this paper, we introduce a new approach for copy-move forgery detection by combining multiscale feature fusion with deep feature extractors, such as Convolutional Neural Networks, CNNs, and image segmentation, to enhance both detection accuracy and localization precision. The idea is to extract features from multiple scales of the image, enabling the system to capture both fine-grained details and larger, global structures that are essential for the identification of tampered regions. This technique fuses features from different scales and applies feature extraction using deep learning. Thus, the technique is used to detect the forgery more accurately even when changes are subtle or the images are low resolution. The method also uses segmentation of images for better localization of the forged regions so that forensic experts will always know which regions were manipulated. Experimental results on publicly available datasets show that our method has a detection accuracy of 94.6% with an improvement of 10-15% over traditional block-based and keypoint-based techniques. Furthermore, our approach shows robustness against noise and compression artifacts, achieving localization precision at 90% for small, highly manipulated regions. These results show the efficiency of the proposed method in real-world forgery detection applications