SGCN: spatially gradient convolution network for certificate document image manipulation localization

Current tampering detection methods pay more attention to natural content images. The research on tampering algorithms for certificate document images is relatively limited, but certificate document images are the most commonly tampered with images, and they cause great harm to society. Our work presents a method for detecting certificate-like image manipulation using the ASGC-Net network. To achieve a network that can better localize text tampering cues. In addition, we propose a spatially constrained convolution that can effectively suppress image content and learn manipulation detection features by capturing different features between the neighborhood and the center of the convolution space. To increase the network's ability to capture tampering cues at multiple scales of images, we add multilayer cross-scale connections inspired by FPN networks. Experiments show that the algorithm is more accurate than general-purpose manipulation detection algorithms in locating tampered regions of certificate document images.