A hybrid Xception-Ensemble model for the detection of Computer Generated images

Abstract

Digital images play a vital role in digital communication due to their applications in various domains like games, movies, and medical and legal spheres. Entities fabricate content through computer-generated images, which causes severe adverse consequences. We propose a novel hybrid Xception-Ensemble approach for distinguishing computer-generated images using the depthwise separable convolution of the Xception architecture. We use depthwise separable convolution and the parameters transferred from the pre-trained ImageNet weights to distinguish the features in computer-generated images with ensemble average learning for efficient classification. The accuracy of the proposed system is better than that of state of the art systems on DSTok, Columbia PRCG and Rahmouni datasets.