Automatic Acquisition and Optimal Application of Illumination Environments Maps

Abstract

The visual quality of digital material appearance representation strongly depends, among others, on representation of realistic illumination conditions. While accurate material appearance can be represented by means of measured illumination and view direction dependent textures (i.e., bidirectional texture function), the illumination conditions can be captured in omnidirectional image (i.e., environment map) that can be, for purpose of fast interactive visualization, represented by an appropriate finite set of point-lights. In this thesis we used two ways of high-dynamic-range environment map acqui- sition obtained either by means of fish-eye photo-lenses, or by taking pictures of mirrored sphere. The main goal of this thesis was to analyse human visual per- ception of three different materials illuminated by ten different types of realistic illumination conditions. We performed two psychophysical experiments with 29 naive volunteers to determine appropriate number of lights necessary to achieve the same visual quality across different materials and illumination types. As a result of data analysis from the experiment we suggested a computationally simple method that can predict the appropriate number of lights for any environment map while still preserving the required visual quality. This method successfully decrease the processing times during rendering and still maintains the realistic visual appearance of any digital representation of real-world materials illumination.