Using rich models to distinguish photorealistic and photographic image

Abstract

Software like Autodesk Maya and Blender have given a significant lift to the photorealism industry. Although an asset for the movies and video game industry and elsewhere also, with advancements in pho torealism, images and videos have lost the stature of genuine record of memories and facts. Attempts are being made to camouflage computer generated photos as photographic images. From the security perspec tives, this raises serious concerns of forging evidence. Consequently, in this work, I propose a system for automatic detection of photorealistic computer generated images (PRCG). The proposed system utilizes rich models for differentiating PRCG from photographic images (PIM). The proposed system achieves close to 99% classification accuracy in many scenarios. Experiments have been conducted on test cases constituting of both single and double JPEG compressed images with varying qual ity factors, something which has not been considered in the existing literature albeit of high practical relevance.