Dynamic Atomic Column Detection in Transmission Electron Microscopy Videos via Ridge Estimation

Area of Science:

• Material Science
• Image Processing
• Computational Physics

Background:

• Transmission Electron Microscopy (TEM) generates image sequences crucial for material analysis.
• Current TEM video analysis often relies on frame-by-frame object recognition, limiting temporal insights.
• Tracking atomic-level objects in TEM requires advanced methods to handle dynamic behavior.

Purpose of the Study:

• To develop advanced ridge detection algorithms for analyzing spatio-temporal image tensors from TEM videos.
• To enable non-parametric estimation of atomic-level object trajectories as a continuous function of time.
• To improve the analysis of TEM image sequences by harnessing temporal correlations across frames.

Main Methods:

• Utilized ridge detection as a classical image processing tool adapted for material science.
• Developed new algorithms for non-parametric trajectory estimation of atomic-level objects.
• Applied spatio-temporal tensor analysis to long image sequences from TEM videos.
• Tailored methods to handle objects with stochastic disappearance and reappearance.

Main Results:

• Demonstrated high effectiveness of the proposed method in simulation scenarios.
• Achieved notable performance improvements in TEM experiments compared to existing benchmarks.
• Successfully estimated explicit trajectories of atomic-level object locations over time.
• Showcased the ability to track objects exhibiting stochastic behavior.

Conclusions:

• The novel ridge detection approach significantly enhances the analysis of TEM image sequences.
• This method provides a more comprehensive understanding of atomic-level object dynamics than traditional techniques.
• The approach offers a powerful new tool for material science research using TEM data.