3D SEM Surface Reconstruction from Multi-View Images

Abstract

The scanning electron microscope (SEM), a promising imaging equipment has been used to determine the surface properties such as compositions or geometries of specimens by achieving increased magnification, contrast, and resolution. SEM micro-graphs, however, remain two-dimensional (2D). The knowledge and information about their three-dimensional (3D) surface structures are critical in many real-world applications. Having 3D surfaces from SEM images provides true anatomic shapes of micro-scale samples which allow for quantitative measurements and informative visualization of the systems being investigated. A novel multi-view approach for reconstruction of SEM images is demonstrated in this research project. This thesis focuses on the 3D SEM surface reconstruction from multi-view images. We investigate an approach to reconstruction of 3D surfaces from stereo SEM image pairs and then discuss how 3D point clouds may be registered to generate more complete 3D shapes from multi-views of the microscopic specimen. Then we introduce a method that uses an algorithm called KAZE, which reconstructs 3D surfaces from multiple views of objects. Then Numerous results are presented to show the effectiveness of the presented approaches.