Double Dual Convolutional Neural Network (D2CNN) for Copy-Move Forgery Detection

Abstract

In recent years, the problem of fake image diffusion is on the rise mainly on social networks because of the development of different tools for image editing. Copy-move forgery (CMF) is one of the image forgeries types used for manipulating the image content. In CMF, the region in an image is copied and placed in a different location in the same image. In this paper, an algorithm for CMF detection based on a Double Dual Convolutional Neural Network (D2CNN) is proposed. A novel concatenation of two Dual Convolutional Neural Networks (DCNN) is used, where each DCNN is composed of two CNN networks. A fully connected network (FCN) is taking the result of the D2CNN and hence classifying the input images into either original or forged. The features extracted from the two DCNN and fusion of these features (D2CNN) have achieved good results according to the following metrics: Accuracy, f1-score, precision, and recall. Two standard datasets namely MICC F-220 and MICC F-2000 have been used to evaluate the proposed approach. Experimental analysis shows that the proposed approach achieves accuracy higher than 98.48% on the MICC F-220 dataset and 97.83% on the MICC F-2000 dataset.