Estimation of illumination spectrum from a hyperspectral image

Abstract

Illumination separation in the red, green, blue (RGB) color space has been an essential topic for color constancy. A simple and non-iterative solution for hyperspectral illumination estimation is proposed in this study, which is based on the assumption that reflectance of natural surfaces can be modeled as low-dimensional linear space. The estimated illumination depends on the null space of the reflectance and the correlation of the hyperspectral image (HSI). This derivation is expected to inspire new theoretical investigations into illumination estimation. © 2019 Institute of Electronics and Information Engineers. All rights reserved.