Optimal Convex reduction of Calibration Targets

Calibration targets are widely used to characterize imaging devices and estimate optimal profiles to map the response of one device to the space of another. The question addressed in this paper is that of how many surfaces in a calibration target are needed to account for the whole target perfectly. To accurately answer this question we first note that the reflectance spectra space is closed and convex. Hence the extreme points of the convexhull of the data encloses the whole target. It is thus sufficient to use the extreme points to represent the whole set. Further, we introduce a volume projection algorithm to reduce the extremes to a user defined number of surfaces such that the remaining surfaces are more important, i.e. account for a larger number of surfaces, than the rest. When testing our algorithm using the Munsell book of colors of 1269 reflectances we found that as few as 110 surfaces were sufficient to account for the rest of the data and as few as 3 surfaces accounted for 86\% of the volume of the whole set.