Image Tampering Detection via Semantic Segmentation Network
Qian Zhao, Gang Cao, Antao Zhou, Xianglin Huang, Lifang Yang · 2020
A large number of image forgery created by image tampering pose a serious threat to the construction of social honesty and content security. There is an urgent requirement to identify tampered images through effective image authentication techniques. In this paper, we propose a new image tampering detection method using the deep learning-based semantic image segmentation network. It works under the observation that tampering regions are typically the semantic object ones in an image. One of the state-of-the-art neural networks for semantic image segment task, namely DeepLab V3+, is used to extract rich and high level features from large amounts of prepared tampering-labeled images. Both spatial regional and boundary tampering artifacts are explored in an encode-decoder network. Experimental results on public image forgery datasets verify the effectiveness of our proposed tampering detection method.