Deep Localization on Mixed Image Tempering Techniques Using U-Net

With advancements in technology, the prevalence of image manipulation for deceptive purposes has increased. Detecting forgeries can pose a challenge, and current methods have limitations. However, computer programs can be utilized to learn from tampered and original images, enabling identification of various types of tampering across a wide range of images. To achieve this, it is crucial to train these programs to locate altered parts within an image. The U-NET deep learning architecture offers a significant improvement in localization accuracy through the use of Error Level Analysis (ELA) and Block Artifact Grid (BAG) error maps as inputs.