Illumination factors such as shading, shadow, and highlight observed from object surfaces affect the appearance and analysis of natural color images. Invariant representations to these factors were presented in several ways. Most of these methods used the standard dichromatic reflection model that assumed inhomogeneous dielectric material. The standard model cannot describe metallic objects. This chapter introduces an illumination-invariant representation that is derived from the standard dichromatic reflection model for inhomogeneous dielectric and the extended dichromatic reflection model for homogeneous metal. The illumination color is estimated from two inhomogeneous surfaces to recover the surface reflectance of object without using a reference white standard. The overall performance of the invariant representation is examined in experiments using real-world objects including metals and dielectrics in detail. The feasibility of the representation for effective edge detection is introduced and compared with the state-of-the-art illumination-invariant methods.
TopIntroduction
Color observed from object surfaces provides crucial information in computer vision and image analysis which include the essential problems of feature detection, image segmentation, object recognition, and image retrieval. However, in real-world applications there are various illumination factors that can affect color images observed from object surfaces. The observed images do not only depend on surface-spectral reflectances and illuminant spectrum, but also include reflection effects such as shading, gloss, and highlight, which mainly depend on illumination geometries and surface materials. Therefore, image representation invariant to shading, shadow, lighting, and highlight was proposed for color image (Slater et al., 1996; Finlayson, 1996; Gevers, 1999; Gevers et al., 2000, 2003; Geusebroek et al., 2000; Smeulders et al., 2001; Geusebroek et al., 2001; Narasimhan et al., 2003; Tan et al., 2005; Park, 2003; Mallick et al., 2005; Van De Weijer et al., 2005, 2006) so far in several ways. These invariant representations play an important role in many applications such as image segmentation (Narasimhan et al., 2003; Park, 2003; Gevers, 2002; Ibrahim et al., 2009a, 2009b, 2010, 2011, 2016; Ali et al., 2015), feature detection, such as edge and corner detection (Geusebroek et al., 2001; Gevers et al., 2000, 2003, 2005; Van De Weijer et al., 2005, 2006; Stokman et al., 2007), object recognition (Gevers et al., 1999, 2004; Van de Sande et al., 2010; Tharwat et al., 2015b), image retrieval (Gevers et al., 2000), cast shadow segmentation (Salvador et al., 2004), optical flow calculation (Mallick et al., 2005: Zickler et al., 2008; Van de Weijer et al., 2004), biometrics (Tharwat et al., 2012a, 2012b, 2015a; Ibrahim et al., 2014, 2015; Gaber et al., 2015, 2016), and robots (Maier et al., 2009, 2010).