Quantifying Photorealism

Computer generated imagery, or CGI, can be used to produce renders of near photographic quality, but what makes these renders considered realistic? Do there exist certain material settings or parameters that make an object more realistic than others? If so, these values can be used to produce incredibly photorealistic renders, and unrealistic parameters can be more easily identified. Prior research in this subject utilized computer algorithms analyzing roughness, color and shadow parameters compared to photographic reference, while another utilized a small sample size of non-artist background human subjects analyzing both photographs and digital renders to determine what material and visual properties quantified photorealism. This paper introduces using a large sample size of human subjects to determine which settings in 3D rendering contribute to perceived realism, using a series of controlled material parameters applied to a selection of CAD components. These material parameters include: metallicity, roughness, specular intensity, and surface detailing intensity and distribution. Preconceived notions towards assuming how realistic a surface material looks can be validated or challenged as a result of these findings. The results of these findings will provide an insight into why certain material settings are perceived as more realistic than others, and will help improve the state of 3D rendering