Computational Imaging for Scene Understanding

5Spectral Modeling and Separation of Reflective-Fluorescent Scenes

Ying FU¹, Antony LAM², Imari SATO³, Takahiro OKABE⁴, and Yoichi SATO⁵

¹Beijing Institute of Technology, China

²Mercari, Inc., Tokyo, Japan

³Computational Imaging and Vision Lab, National Institute of Informatics, Tokyo, Japan

⁴Department of Artificial Intelligence, Kyushu Institute of Technology, Fukuoka, Japan

⁵Institute of Industrial Science, The University of Tokyo, Japan

5.1. Introduction

Hyperspectral reflectance data are beneficial to many applications including but not limited to archiving for cultural e-heritage (Balas et al. 2003), medical imaging (Styles et al. 2006) and also color relighting of scenes (Johnson and Fairchild 1999). As a result, many methods for acquiring the spectral reflectance of scenes have been proposed (Maloney and Wandell 1986; Tominaga 1996; Gat 2000; DiCarlo et al. 2001; Park et al. 2007; Chi et al. 2010). Despite the success of these methods, they have all made the assumption that fluorescence is absent from the scene. However, fluorescence does frequently occur in many objects, such as natural gems and corals, fluorescent dyes used for clothing and plant containing chlorophyll to name a few. In fact, Barnard shows that fluorescent surfaces are present in 20% of randomly constructed scenes (Barnard 1999). This is a significant proportion of scenes that have not been considered by most of the past methods.

Figure 5.1. (a) The scene captured under white light. (b) The recovered ...