Identifying natural images and computer-generated graphics based on Convolutional Neural Network

Abstract

Aiming at the identification of natural images and computer-generated graphics, an image source pipeline forensics method based on convolutional neural network (CNN) is proposed. In this method, Inception-v3 is used as the basic network, and the pre-trained model parameters in ImageNet are adopted. The top-level classification layer of Inception-v3 is replaced by two fully-connected Softmax classifiers. With the transfer learning, a new network model is constructed. The network is fine-tuned by a database with 10,000 images to identify natural images and computer-generated graphics. Experimental results and analysis show that it can effectively identify natural images and computer-generated graphics, and it is robustness against JPEG compression, scaling, rotation, noise and other post-processing operations. Furthermore, the effect of Softmax classifier and SVM classifier on the experimental results are analysed.