A Novel Face Forgery Detection Method Based on Edge Details and Reused-Network

Abstract

In recent years, with the widespread popularity of digital media editing software and rapid development of deep learning techniques, the technology barriers for generating fake faces becomes lower and lower. As a result, forged face images have become more and more realistic, and a large number of high-quality fake face images are created and spread in our day lives. Therefore, determining the authenticity of face images is a challenging task and major focus of multimedia security research. To solve this problem, a novel face forgery detection method based on edge details and Reused-Network (RN) is proposed. Specifically, based on Sobel and Laplacian operators, we consider the differences in edge details between real and fake faces, and further learn these differences through RN to eventually extract more detailed features. The derived features are then used to train a Support Vector Machine (SVM) classifier for classification. The proposed method has a good performance on detecting fake face images, and it is experimentally verified better than some state-of-the-art works.