Polarization-Guided 3-D Reconstruction for Occluded Faces Using Passive Millimeter-Wave Single-Direction Imaging

Due to its ability to penetrate materials such as masks, beards, and clothing, passive millimeter-wave (PMMW) imaging can be employed in occluded face recognition (OFR). The 3-D structure, as an important feature that can enhance recognition capabilities, has always been a challenge in the acquisition of information in PMMW imaging. Polarization imaging can effectively characterize the target surface information. This article proposes a new physics-based 3-D reconstruction method based on single-direction PMMW polarization imaging. The surface normal vector (SNV) of the target is inverted using polarization features, specifically the DoLP and the angle of polarization (AoP). Moreover, a model to resolve the azimuth ambiguity problem in centrosymmetric or approximately centrosymmetric targets is presented without any prior information. The feasibility of the proposed method is validated through simulation experiments, as well as measurement experiments on spheres and human faces with various occlusions. Contour and performance analyses demonstrate that the method is capable of reconstructing 3-D structures of the target and can penetrate occlusions to reconstruct local facial structural features. Comparison to existing methods further shows that our method achieves better results and holds promise for assisting OFR.