A Deep Learning Approach for Image Forgery Detection and Localization Using UNet Segmentation
Main Article Content
Abstract
Image forging, a common kind of digital alteration, jeopardizes the validity and reliability of visual output. Detecting such forgeries has become a crucial challenge in a variety of fields, including media, law enforcement, and digital forensics, due to the fast evolution of image altering tools and methods. In this paper, we offer a unique method for detecting picture counterfeiting using U-Net segmentation, a deep learning architecture well-known for its performance in image segmentation tasks. The U-Net design was selected because of its capacity to capture fine features and spatial connections within a picture. We modify and expand the U-Net architecture to detect forgeries using segmentation, with the objective of identifying modified parts inside an image. Our method makes use of a collection of legitimate and fabricated photos, allowing the U-Net model to learn distinguishing characteristics that distinguish between the two. The suggested technique entails training the U-Net model on a wide collection of photos, which includes image forgeries such as copy-move, splicing, and retouching. The U-Net learns to segment a picture into multiple sections during training, therefore learning the intrinsic patterns and textures indicative of manipulation. Following that, the trained model is used to examine fresh, previously viewed photos by highlighting and localizing locations suspected of fabrication. The experimental findings illustrate the efficacy of our U-Net-based forgery detection method. In comparison to current forgery detection systems, it outperforms them in terms of accuracy, precision, and recall. Furthermore, we test the model's resistance to various counterfeit tactics and noise levels, ensuring its dependability in real-world circumstances.