Fast Spectral Upsampling of Volume Attenuation Coefficients

Abstract

Most of the input data for rendering is authored in tristimulus colors. In a spectral renderer these are usually converted into spectra that have more than three degrees of freedom while maintaining the authored appearance under a white illuminant. For volumes there are two additional problems: (1) the nonlinear color shift over the distance traveled through the volume and (2) the large amount of data or procedural data. Problem 1 makes it difficult to preserve the appearance, whereas problem 2 makes it infeasible to precompute the spectral data. This chapter shows that using box-like spectra produces the closest fit to a tristimulus render of the tristimulus data. We propose to optimize two threshold wavelengths depending on the input color space that are then used to convert the volume coefficients on the fly.