Digital Image Forgery Detection Based on Convolutional Neural Networks

Image authentication has become a hot topic with the currently available technology for manipulating and distributing images. Image authentication aims to ensure the authenticity of digital images and automatically detect forged images that have been tampered with after they have been captured. This paper presents a convolutional network (CNN) model for detecting forged images. The proposed model implies three main stages: First, preprocessing the input image by adopting logarithmic mapping to refine the quality of the extracted features, especially in dark regions. Secondly, a novel CNN architecture was trained to classify arbitrary images into two categories: “original” and “forged”. The CNN model locates some descriptors as a descriptor map. Finally, the model finds the similarities and dependencies between these features. The proposed model was tested and evaluated on three datasets under various copy-move conditions. Experimental results reveal that the model can detect forged images with high accuracy, reaching up to 97.11%.