Computational Imaging for Scene Understanding

11Polarization Imaging in the Wild Beyond the Unpolarized World Assumption

Jérémy Maxime RIVIERE

Realistic Graphics and Imagining, Imperial College London, UK

11.1. Introduction

Polarization imaging has been widely studied in the past for material classification (Wolff 1990), shape estimation (Wolff 1989; Miyazaki and Ikeuchi 2005; Guarnera et al. 2012; Kadambi et al. 2015; Smith et al. 2016) and reflectometry (Miyazaki et al. 2003). All of these methods make the “unpolarized world” assumption, i.e. they assume the incident illumination is completely unpolarized. The common guiding thread of all these methods is to measure the polarization induced by reflection off the material’s surface, by taking multiple photographs (at least three) of the object through a rotating linear polarizer placed in front of the camera sensor. In this chapter, we will first cover some background on polarization from specular reflection and the necessary mathematical tools to handle polarized light. We continue to follow the assumption that diffuse reflection completely depolarizes the incident illumination. While this has been shown not to hold for 3D objects near occluding contours (Atkinson and Hancock 2006), it greatly simplifies the foundational theories for the case covered in this chapter: capturing the appearance of spatially varying, planar surfaces under unconstrained, outdoor illumination.

Figure 11.1. Plots of the Fresnel equations and degree of polarization at an air–dielectric interface. ...