Computational Imaging for Scene Understanding

12Multispectral Polarization Filter Array

Kazuma SHINODA

Graduate School of Regional Development and Creativity, Utsunomiya University, Japan

12.1. Introduction

Imaging of the spectral or polarization spatial distributions of objects is expected to have various applications in the field of computer vision. Although various methods have been studied for taking multispectral and polarization images, the optical systems employed to date have been too large and the methods require capturing the image several times. Filter arrays that transmit different wavelengths and polarization for each pixel have been considered as a method to capture images with a single exposure. Some representative examples are shown in Figure 12.1. A 5-band multispectral filter array (MSFA) with RGB plus cyan and orange has been developed, and a method to reconstruct multispectral images from single exposure images (i.e. demosaicking) has been proposed by Monno et al. (2015). Jia et al. (2016) proposed a 16-channel MSFA with sinusoidal sensitivity using a Fabry–Perot type filter (Jia et al. 2016), called a Fourier spectral filter array, which can remove aliasing. Gruev et al. (2010) also proposed a filter array method for capturing polarization images, where linear polarizers of 0, 45, 90 and 135° are periodically arranged in each pixel to visualize the polarization intensity and angle (Gruev et al. 2010). Sony’s IMX250MYR is a method to achieve RGB and polarization images on the same imager (Sony Semiconductor ...