A new connection space for low-dimensional spectral color management

Multi- or hyper-spectral pixels are usually represented as vectors with high dimensionality. For many applications, not all of these dimensions are actually necessary, and a few values are enough to measure and/or process the pixel with a very good accuracy. In this work, we introduce a new strategy to reduce the dimensionality of spectral images ranging in the visible wavelengths, for purposes of color management. We define a new Interim Connection Space (ICS) that contains only five dimensions, and show that it has numerous advantages over state-of-the-art ICS such as LabPQR. In particular, it allows for a better spectral reconstruction accuracy.