How Real is Really? A Perceptually Motivated System for Quantifying Visual Realism in Digital Images

Contemporary video games are generally designed with the aim of engendering a state of immersion in their players. A sense of ´presence´ is considered to be a prerequisite of total immersion, the ultimate level of immersion, and can be directly related to the degree of realism in the simulation presented in a videogame. However, continuously increasing the degree of visual realism in an image could cause adverse effects such as fault amplification and more importantly, emotions like fear and dread of the uncanny. It can also substantially raise the development costs of a game. Consequently, videogame developers would benefit greatly from knowing the threshold beyond which increased realism is counter-productive. This paper details the development of an image processing system that measures the degree of realism of an image using three methods - gradient variance, color variance and shadow softness - derived from existing theories and practices in perceptual psychology, photoforensics and image processing. It then discusses testing of the system on a variety of photographic and video game images.