Automated surface profile measurement with digital holographic microscopy and large mask inpainting networks

Abstract

Digital holographic microscopy (DHM) allows for highly precise 3D surface measurements in a non-invasive way, but phase aberrations from off-axis DHM recordings can compromise image accuracy. Traditional compensation methods require manual intervention, hindering further automated use of DHM. Other methods based on background segmentation and Zernike polynomials have been proposed, but identifying the sample and background regions can lead to inaccurate compensation results. Additionally, traditional image restoration algorithms struggle with restoring sample-free holograms involving large or multiple vacant areas. A new automated aberration compensation method is proposed using large-mask inpainting networks. This method restores sample-free holograms and compensates for phase aberrations, leveraging deep learning to enable real-time measurements. In the study, the network was trained with holograms of varying fringes, and experimental results show its effectiveness in improving image accuracy and detail. This approach could have wide applications in industries such as micro-electromechanical systems and micro- integrated circuits.