A Comparison of Two Machine Learning Approaches for Photometric Solids Compression

Abstract

The use of photometric solids into both real time and photorealistic rendering allows designers and computer artists to enhance easily the quality of their images. Lots of such data are available from lighting societies since they allow these societies to easily present the luminance distribution of their often complex ligthing systems. When accuracy is required the amount of discretized luminance directions and the number of photometric solids that have to be used increase considerably the memory requirements and reduce the algorithm efficiency. In this paper we describe and compare two machine learning approaches used for approximating any photometric solid: an artificial neural network and ECON (Equi-Correlated Network Algorithm). By applying these two approaches on a large set of real photometric distribution data, we were able to show that one of them provides generally a better approximation of the original distribution.