For textureless objects with smooth surfaces in vision tasks, conventional three-dimensional reconstruction methods cannot retrieve the valid shape information of surfaces due to the loss of structure-texture features and the high reflectance.This paper presents a 3D reconstruction algorithm for textureless objects based on optical polarization imaging, which does not rely on the structure-texture features of objects.Stokes vector is calculated to quantify the polarization states of the surface reflected light.Then, based on the analysis of polarimetric-geometric space, the normal distribution of surface is effectively estimated.Finally, the multi-scale Shapelets blank operator is applied to integrate the normal vector to derive the accurate depth information of the objects.Experiments show that the proposed method can obtain the correct 3D shape reconstruction results, especially for high-reflective objects with less texture features.Meanwhile, the proposed technique can suppress the specular highlight and noise, with high computational efficiency.