Feature Pyramid Network for Diffusion-Based Image Inpainting Detection

Abstract Inpainting is a technique that can be employed to tamper with the content of images. In this paper, we propose a novel forensics analysis method for diffusion-based image inpainting based on a feature pyramid network (FPN). Our method features an improved u-shaped net to migrate FPN for multi-scale inpainting feature extraction. In addition, a stagewise weighted cross-entropy loss function is designed to take advantage of both the general loss and the weighted loss to improve the prediction rate of inpainted regions of all sizes. The experimental results demonstrate that the proposed method outperforms several state-of-the-art methods, especially when the size of the inpainted region is small.