Three-dimensional reconstruction of super-resolved white-light interferograms based on deep learning

White-light scanning interferometry is an effective and widely used technology for measuring the microscopic three-dimensional morphology of an object. Its vertical resolution can reach the sub-nanometer level, and its lateral resolution reaches submicron level. However, for the samples containing complex structure or high-density periodic distribution structural units, the measurement results are strongly restricted by magnification and numerical aperture (NA) of the microscopic objective. In this paper, we proposed a three-dimensional reconstruction algorithm for white-light interferograms after super resolution processing, using fast super-resolution convolutional neural networks (FSRCNN) to improve the detailed information of the interferograms, and then we used centroid method combined with the five-step phase-shift method to extract the zero optical path difference (ZOPD) position of the interference signal after super resolution processing. After processed by the proposed method, the interferograms collected by the 10X microscope objective (NA= 0.3) recovered the 3D surface is the same as that measured by the 100X microscope objective (NA= 0.7), which is proved by the experiment results.