Exploring visual and automatic measures of perceptual fidelity in real and simulated imagery
Academic Article
Overview
Research
Identity
Additional Document Info
Other
View All
Overview
abstract
This paper introduces a new psychophysical experiment developed to enable observers to judge the quality of computer graphics imagery with respect to the real scene it depicts. This new framework facilitates perceptual judgment of images against a real scene. Unlike previous work, which examined primitive objects under basic illumination, this experiment examines complex geometry illuminated using a calibrated light source. To ensure valid results, a commercial lighting booth containing rapid prototyped three-dimensional (3D) objects serves as the real scene. For comparison, a series of representative images, of varying quality, were rendered using the physically based Radiance lighting simulation software. Results from these experiments show that higher parameter settings, which lead to longer processing times, do not necessarily lead to higher quality images. To demonstrate that there is only modest benefit to setting parameters higher, images are subjected to further testing using two different visual quality discrimination operators; the Visual Differences Predictor (VDP) and the Structural SIMilarity (SSIM) for image-quality assessment. The results from the automatic operators correspond well with each other, in addition to yielding comparable outcomes as the psychophysical experiment. Although, a single scene was considered in the experiment, several scenes are tested using the image-quality metrics to lend further reliability to the assertion that higher parameter settings, which lead to extended processing times, do not necessarily lead to superior quality results.
