SD-SEM: sparse-dense correspondence for 3D reconstruction of microscopic samples.

Area of Science:

• Materials Science
• Biomedical Engineering
• Mechanical Engineering

Background:

• Scanning electron microscopy (SEM) provides high-resolution 2D images but lacks depth information.
• 3D reconstruction from stereo SEM images is crucial for detailed analysis in various scientific fields.

Purpose of the Study:

• To develop and investigate a novel framework for 3D reconstruction of stereo SEM images.
• To enable depth recovery from 2D SEM micrographs.

Main Methods:

• Utilized stereo SEM imaging by tilting the specimen stage.
• Employed sparse scale-invariant feature transform (SIFT) for image rectification and outlier removal.
• Applied dense SIFT descriptors, factor graph energy minimization, and loopy belief propagation (LBP) for dense correspondence estimation and depth recovery.

Main Results:

• Successfully achieved pixel-by-pixel correspondence between stereo SEM images.
• Demonstrated high-quality 3D reconstruction of microscopic samples.
• Validated the effectiveness of the sparse-dense correspondence framework.

Conclusions:

• The proposed framework offers a robust solution for 3D reconstruction from stereo SEM data.
• This method enhances the analytical capabilities of SEM imaging by providing depth information.
• The technique is suitable for detailed analysis in biomedical, mechanical, and materials science applications.