Towards a perceptual quality metric for computer-generated images

The physical validation of computer-generated images (CGIs) has received a lot of attention from the computer graphics community, as opposed to the assessment of these images’ psychovisual quality. The field indeed lacks the proper tools to quantify the perceptual quality of a CGI; this paper engages in the construction of such a metric. A psychovisual experiment was submitted to a representative panel of observers, where participants were asked to score the overall quality and aspects of this quality for several CGIs. An analytical quality model, fit to the data, next gives insight into the relative perceptual importances of these aspects. Accuracy in the simulation of shadows, good contrast and absence of noise were found to have a major impact on the perceived quality, rather than precise anti-aliasing and faithfull color bleeding.