Subsampling and inpainting approaches for electron tomography.

Area of Science:

• Materials Science
• Microscopy
• Image Processing

Background:

• Electron tomography (ET) is crucial for 3D nanoscale imaging.
• Sample damage during data acquisition limits ET resolution and utility.
• Existing reconstruction methods struggle with limited or noisy projection data.

Purpose of the Study:

• To develop novel reconstruction strategies for electron tomography to mitigate sample damage.
• To introduce improved inpainting and subsampling techniques for ET data.
• To enhance the accuracy and efficiency of 3D reconstructions from limited projection data.

Main Methods:

• Proposed higher-order total-variation (HOTV) inpainting as an alternative to standard total-variation (TV) inpainting for projection images.
• Extended HOTV inpainting to 3D, utilizing both image and sinogram data.
• Introduced gradient subsampling as a more efficient data acquisition strategy compared to random subsampling.

Main Results:

• Demonstrated that HOTV inpainting is more suitable for smooth projection images than TV inpainting.
• Showcased the effectiveness of the 3D HOTV extension for comprehensive reconstruction.
• Proved that gradient subsampling is more efficient than random subsampling.
• Rigorous comparison showed new schemes perform comparably or better than existing methods, especially at low subsampling rates.

Conclusions:

• The proposed HOTV inpainting and gradient subsampling strategies offer significant improvements for electron tomography.
• These methods effectively reduce the impact of sample damage by optimizing data acquisition and reconstruction.
• The developed techniques enhance the feasibility of high-resolution 3D imaging with electron tomography, particularly under dose-limited conditions.