Image Forgery Detection Based on Motion Blur Estimated Using Convolutional Neural Network

Abstract

Currently images are key evidences in many judicial or other identification occasions, and image forgery detection has become a research hotspot. This paper proposes a novel motion blur based image forgery detection method, which includes three steps. First, a convolutional neural network (CNN)-based motion blur kernel reliability estimation method is proposed, which is used to determine whether an image patch should be involved in the image forgery detection process. Second, a shared motion blur kernels-based image tamper detection method is proposed to detect whether a group of motion blur kernels are projected from the same 3D camera trajectory effectively. Third, a consistency propagation method is proposed to localize tampered regions efficiently. Experiments on synthetic images and natural images show the availability of the proposed method.