Yuffa Alex J, Gurton Kristan P, Videen Gorden
Appl Opt. 2014 Dec 20;53(36):8514-21. doi: 10.1364/AO.53.008514.
We use a polarimetric camera to record the Stokes parameters and the degree of linear polarization of long-wavelength infrared radiation emitted by human faces. These Stokes images are combined with Fresnel relations to extract the surface normal at each pixel. Integrating over these surface normals yields a three-dimensional facial image. One major difficulty of this technique is that the normal vectors determined from the polarizations are not unique. We overcome this problem by introducing an additional boundary condition on the subject. The major sources of error in producing inversions are noise in the images caused by scattering of the background signal and the ambiguity in determining the surface normals from the Fresnel coefficients.
我们使用偏振相机来记录人脸发出的长波红外辐射的斯托克斯参数和线偏振度。这些斯托克斯图像与菲涅耳关系相结合，以提取每个像素处的表面法线。对这些表面法线进行积分可得到三维面部图像。该技术的一个主要困难在于，由偏振确定的法线向量不是唯一的。我们通过在对象上引入额外的边界条件来克服这个问题。产生反演误差的主要来源是背景信号散射导致的图像噪声，以及从菲涅耳系数确定表面法线时的模糊性。
