R. Schettini, S. Tominaga, and A. Trémeau (Eds.): CCIW 2011, LNCS 6626, pp. 85–98, 2011. © Springer-Verlag Berlin Heidelberg 2011 Estimation of Multiple Illuminants Based on Specular Highlight Detection Yoshie Imai 1,2 , Yu Kato 1 , Hideki Kadoi 1 , Takahiko Horiuchi 1 , and Shoji Tominaga 1 1 Graduate School of Advanced Integration Science, Chiba University, Japan 2 Toshiba Corporation {yoshie.imai,yu_kato,kadoi}@graduate.chiba-u.jp, {horiuchi,shoji}@faculty.chiba-u.jp Abstract. This paper proposes a method for estimating the scene illuminant spec- tral power distributions of multiple light sources under a complex illumination environment. The spectral power distributions including natural and artificial il- luminants are estimated based on the image data from a high-dimensional spectral imaging system. We note that specular highlights on inhomogeneous dielectric object surfaces includes much information about scene illumination according to the dichromatic reflection model. First, we describe several methods for detecting specular highlight areas. We assume a curved object surface illuminated by multi- ple light sources from different directions. Then we estimate the illuminant spec- trum of each light source from the image data of that highlight area. Based on this principle, we present an algorithm to estimate multiple illuminants. The feasibility of the proposed method is shown in experiments. Keywords: Multiple light sources, dichromatic reflection model, specular highlight area, illuminant estimation. 1 Introduction Estimation of scene illumination from image data has important imaging applications, including illumination design, color constancy, image processing, image rendering, and image retrieval. The scene illuminant estimation problem has a long history. In the past, many algorithms were proposed for scene illuminant estimation; however most algorithms assumed uniform illumination from a single light source [1-10]. It should be noted that our illumination environment is not necessarily a single light source such as daylight or a light bulb, but often consists of multiple light sources from different directions. Recently, artificial light sources such as various fluorescent lights and light emit- ting diode (LED) are now in use in daily life. A complex illumination environment is constructed by the multiple illuminants including these artificial light sources and natural daylight. This paper proposes a method for estimating the scene illuminant spectral power dis- tributions of multiple light sources under a complex illumination environment. A high- dimensional spectral imaging system is first realized using a liquid-crystal tunable filter,