Supplemental Materials : Efficient Multispectral Facial Capture With Monochrome Cameras

CIC(2018)

引用 0|浏览6
暂无评分
摘要
1 METHOD: MULTISPECTRAL POLARIZATION PROMOTION We develop a technique that we call multispectral polarization promotion in which we hallucinate cross-polarized images for each spectral channel from unpolarized lighting images, so that we can generate a multispectral diffuse albedo texture map of the subject. Our process requires that only one of the spectral channels in the lighting rig is polarized in the pattern of Ghosh et al. [2011]. For clarity, we extend the variable naming conventions of Ma et al. [2007]. We define a gradient illumination image of the subject Ll,i,s , where l describes the gradient condition, i describes the polarization state (one of cross or parallel), and s defines the index of spectrum of illumination, ranging from 0 to n − 1 where n is the number of spectral channels in the lighting rig, and 0 represents the white LED. The gradient illumination images required [Ma et al. 2007] are therefore: • Lx,c,0, cross-polarized, x gradient • Ly,c,0, cross-polarized, y gradient • Lz,c,0, cross-polarized, z gradient • Lf ,c,0, cross-polarized, full sphere • Lx,p,0, parallel-polarized, x gradient • Ly,p,0, parallel-polarized, y gradient • Lz,p,0, parallel-polarized, z gradient • Lf ,p,0, parallel-polarized, full sphere When linear polarizers over the light sources are oriented perpendicularly to the those in front of the camera, the polarizer will block all of the specularly-reflected light and about half of the diffusely reflected light, such that Ll,i=c,s = 2Dl,s , representing an image of the diffuse or sub-surface scattered reflections. When
更多
查看译文
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要