Institute of Measurement and Automatic Control, Faculty of Mechanical Engineering, Leibniz University Hannover, Nienburger Str. 17, 30167 Hannover, Germany.
Sensors (Basel). 2020 Jun 26;20(12):3598. doi: 10.3390/s20123598.
Scanning electron microscopes (SEMs) are versatile imaging devices for the micro- and nanoscale that find application in various disciplines such as the characterization of biological, mineral or mechanical specimen. Even though the specimen's two-dimensional (2D) properties are provided by the acquired images, detailed morphological characterizations require knowledge about the three-dimensional (3D) surface structure. To overcome this limitation, a reconstruction routine is presented that allows the quantitative depth reconstruction from SEM image sequences. Based on the SEM's imaging properties that can be well described by an affine camera, the proposed algorithms rely on the use of affine epipolar geometry, self-calibration via factorization and triangulation from dense correspondences. To yield the highest robustness and accuracy, different sub-models of the affine camera are applied to the SEM images and the obtained results are directly compared to confocal laser scanning microscope (CLSM) measurements to identify the ideal parametrization and underlying algorithms. To solve the rectification problem for stereo-pair images of an affine camera so that dense matching algorithms can be applied, existing approaches are adapted and extended to further enhance the yielded results. The evaluations of this study allow to specify the applicability of the affine camera models to SEM images and what accuracies can be expected for reconstruction routines based on self-calibration and dense matching algorithms.
扫描电子显微镜（SEM）是一种多功能的微纳成像设备，广泛应用于生物学、矿物学或机械学等多个领域，用于对各种样本进行特征分析。尽管采集的图像提供了样本的二维（2D）特性，但详细的形态特征需要了解三维（3D）表面结构。为了克服这一局限性，提出了一种重建程序，允许从 SEM 图像序列中进行定量深度重建。基于 SEM 的成像特性，可以很好地用仿射相机来描述，所提出的算法依赖于使用仿射极线几何、通过因子分解和密集对应关系的三角测量进行自校准。为了获得最高的鲁棒性和准确性，将仿射相机的不同子模型应用于 SEM 图像，并将获得的结果与共聚焦激光扫描显微镜（CLSM）的测量结果直接进行比较，以确定理想的参数化和基础算法。为了解决仿射相机立体对图像的校正问题，以便应用密集匹配算法，对现有的方法进行了改进和扩展，以进一步提高结果的质量。本研究的评估允许指定仿射相机模型在 SEM 图像中的适用性，以及基于自校准和密集匹配算法的重建程序可以预期的准确性。
