Performance Comparison of Deep Learning Models for Computer Generated Image Detection
C S Sychandran, R. Shreelekshmi · 2023
With the advancement in digital image processing technology, there is a huge proliferation of computer-generated images. They are widely used in interactive games, film industry and cartoons. Computer-generated images are photo-realistic because they are created using complex algorithms and high-end graphics processing power to accurately mimic real-world objects and environments’ lighting, textures, and colors. Since computer-generated images are more realistic, even human eyes can barely distinguish them. So verifying image authenticity and differentiating computer-generated images from photographic images are inevitable. This paper investigates the performance of different deep learning models for detecting computer-generated images. We experiment on three datasets: DSTok, Columbia PRCG and Rahmouni. Experimental results show that the hybrid Xception ensemble outperforms the state-of-the-art methods and achieves the highest accuracy, precision, recall, F1 score and specificity on three benchmark datasets.