Abstract
This paper proposes a method for estimating the scene illuminant spectral power distributions of multiple light sources under a complex illumination environment including natural and artificial illuminants. We assume that curved object surfaces of dielect materials are illuminated by the multiple light sources from different directions. Light reflection from the surfaces is described by the dichromatic reflection model, in which the specular reflection component includes much information about scene illuminant. First, we describe a high-dimensional spectral imaging system. Second, we propose several methods for detecting the candidates of specular highlight areas in the captured image, and determine the most proper highlight areas on the object surfaces. Third, we present an estimation algorithm of illuminant spectra from the detected highlight areas. All the estimated illuminant spectra are summarized into some classes based on a similarity measure. The feasibility of the proposed method is shown in experiments using real objects under several light sources.