Polarization is useful for estimating surface normals. In this paper, polarization is applied to shape estimation. Surface normal estimation from a single image is commonly performed by shape-from-shading. An alternative approach is shape-from-polarization, which usually cannot uniquely determine the surface normal and must be accompanied by additional information. The shading information is a promising option, but if the index of refraction is unknown, the surface normal is still non-uniquely determined. To resolve this problem, we assume that the object surface is locally convex. Under this assumption, we show that the surface normal can be uniquely determined from the gradients of the shading information. Our method works when the light and camera are closely arranged.

极化对于估计表面法线很有用。在本文中，极化应用于形状估计。来自单个图像的表面法线估计通常通过着色形状来执行。另一种方法是极化形状，它通常不能唯一地确定表面法线，必须附有附加信息。着色信息是一个很有前景的选项，但如果折射率未知，表面法线仍然是非唯一确定的。为了解决这个问题，我们假设物体表面是局部凸面的。在这个假设下，我们表明可以从着色信息的梯度唯一地确定表面法线。当灯光和相机紧密排列时，我们的方法有效。