TrueEBSD in MTEX: automatic image matching for correlative microscopy applications

This paper introduces TrueEBSD, an open-source MATLAB add-on for MTEX that enables automatic image alignment and spatial distortion correction to facilitate correlative microscopy analysis by integrating EBSD maps with other imaging data for enhanced quantitative crystallographic measurements.

The Gist

Imagine you are trying to solve a giant jigsaw puzzle, but the pieces you have are from two different boxes. One box contains a map of the puzzle's "skeleton" (showing the shapes and directions of the pieces), and the other box contains a photo of the puzzle's "surface" (showing the colors and textures).

The problem is that these two images were taken at slightly different times and angles. Because of this, the "skeleton" map is stretched, tilted, or shifted compared to the "surface" photo. If you try to lay them on top of each other, the edges don't match, and the picture looks blurry and wrong. You can't get a true understanding of the puzzle because the two views don't line up.

This is exactly the problem scientists face when studying materials like metal alloys or copper. They use two powerful tools:

1. EBSD: A microscope technique that maps the internal "crystal skeleton" of a material (how the atoms are arranged).
2. SEM Imaging: A standard microscope photo that shows the surface texture, cracks, or different material phases (like a black-and-white photo vs. a color photo).

Usually, these two images don't line up perfectly due to tiny shifts, tilts, or drifts in the microscope.

The Solution: TrueEBSD

The paper introduces a new software tool called TrueEBSD (now built into a popular toolbox called MTEX). Think of TrueEBSD as a smart, automatic "glue" and "straightener" for these mismatched images.

Instead of a human having to manually pick points to line up the images (which is slow and prone to human error), TrueEBSD does the work automatically. It looks for common features in both images—like the edges of grains or specific patterns—and calculates exactly how much one image needs to be stretched, shifted, or tilted to match the other.

How it works in simple steps:

1. It takes a stack of images: It starts with the most distorted image and works its way to the "perfect" reference image.
2. It measures the wobble: It breaks the image into small chunks and measures how much each chunk has moved relative to the others.
3. It fixes the math: It uses mathematical models to smooth out these movements, effectively "warping" the distorted image until it fits perfectly over the reference image.
4. It creates a super-map: Once aligned, it combines the data. You now have a single map that shows both the internal crystal structure and the surface features in perfect registration.

Real-World Examples from the Paper

The authors tested this "digital glue" on two specific materials to show how powerful it is:

1. The "Hard Metal" Puzzle (WC-Co Composites)

• The Material: A mix of hard tungsten carbide (WC) grains stuck together by a cobalt (Co) binder. This is used for cutting tools.
• The Problem: The microscope used to map the crystals (EBSD) is bad at seeing the cobalt binder. It often thinks there is less cobalt than there really is, like a blurry photo missing details. This leads to wrong calculations about how tightly the hard grains are packed together.
• The Fix: TrueEBSD aligned the blurry crystal map with a sharp, high-contrast photo of the surface. It then "painted" the correct cobalt areas onto the crystal map.
• The Result: Scientists could finally measure exactly how much cobalt was there and how the hard grains touched each other, giving a much more accurate picture of the material's strength.

2. The "Copper" Puzzle (Grain Boundaries and Voids)

• The Material: A block of copper metal.
• The Problem: Under stress, tiny holes (voids) form in the copper, usually along the boundaries where different crystals meet. Scientists want to know: Do these holes form at random, or do they avoid certain types of boundaries?
• The Fix: They aligned the crystal map with a photo showing the tiny holes. Because the images were now perfectly overlaid, they could see exactly which type of crystal boundary a hole was sitting on.
• The Result: They discovered that a specific type of boundary (called a "Sigma 3 twin boundary") acts like a shield—it rarely gets holes. Other boundaries, however, are vulnerable. This helps engineers design copper that lasts longer.

Why This Matters

Before this tool, scientists had to do this alignment manually, which was tedious and subjective (different people might get different results). TrueEBSD automates the whole process. It's like upgrading from hand-drawing a map to using a GPS that automatically corrects for traffic and road shifts.

The paper emphasizes that this tool is open-source (free for everyone to use), fast (it uses clever coding tricks to run quickly), and flexible (it can handle all sorts of different microscope setups). By making these images line up perfectly, it allows scientists to ask and answer questions that were previously impossible to solve because the data was too messy to combine.

Technical Summary

1. Problem Statement

Correlative microscopy, which combines data from multiple imaging modalities (e.g., Electron Backscatter Diffraction (EBSD) and Scanning Electron Microscopy (SEM)), is essential for comprehensive materials characterization. However, a significant barrier to effective correlative analysis is spatial misalignment caused by distortions between different imaging modes.

• Sources of Distortion: These include specimen tilt, temporal drift, electromagnetic field shifts, and rolling-shutter effects.
• Consequences: Without precise pixel-wise alignment, data cannot be overlaid accurately. This prevents the augmentation of crystallographic maps with complementary data (e.g., phase fractions or topography), leading to systematic errors.
  • Example: EBSD systematically underestimates the volume fraction of low atomic-number phases (like Cobalt in WC-Co composites) due to asymmetric spatial resolution.
  • Example: Identifying void nucleation sites on specific grain boundaries requires aligning voids (visible in BSE images) with crystallographic boundaries (visible in EBSD), which is impossible if the maps are misaligned.
• Limitations of Previous Tools: The original TrueEBSD MATLAB program required manual intervention (control points) or was rigid in its workflow, limiting automation and flexibility for large or complex datasets.

2. Methodology

The authors re-implemented the TrueEBSD algorithm as a modular, open-source add-on for MTEX, a widely used MATLAB toolbox for EBSD data analysis. The new implementation (v2.1.0) introduces a class-based workflow and several technical enhancements.

A. Algorithmic Workflow

The process aligns a sequence of images (or maps) to a final reference image by measuring and correcting local shifts:

1. Input: A list of distorted images/maps (e.g., EBSD band contrast, FSE, BSE).
2. Resampling: Images are resampled to a common pixel grid using linear interpolation (for images) or nearest-neighbor (for categorical data like orientations).
3. Cross-Correlation: The algorithm uses Fast-Fourier Transform (FFT)-based cross-correlation on Regions of Interest (ROIs) at image edges to measure local shift vectors.
4. Model Fitting: Measured shifts are fitted to mathematical models based on physical distortion types:
   • Affine Transform: For electromagnetic field shifts.
   • Projective Transform: For specimen tilt.
   • Linear Spline: For rolling-shutter drift.
5. Correction: Images are warped using the fitted model to align them pixel-wise with the reference image.

B. Key Technical Innovations

• Class-Based Architecture: The software is structured around three main classes:
  • @distortedImg: Stores image sequences and metadata.
  • @pairShifts: Stores displacement fields and ROI parameters.
  • @trueEbsd: The main workflow object managing the correction steps.
  • Benefit: This allows for easy integration with MTEX objects (e.g., storing EBSD orientations alongside images) and reusability across different experimental setups.
• Automatic ROI Optimization: Instead of fixed ROI sizes, the algorithm iteratively adjusts the ROI width. If the residual error after alignment exceeds 2 pixels, the ROI size is doubled. This balances the need for feature-rich correlation against the "exclusion zone" at image borders.
• Performance Optimization: The computationally intensive cross-correlation function was rewritten as MEX files (C binaries compiled from MATLAB code). This removes the dependency on the Parallel Computing Toolbox while significantly increasing processing speed.
• Graphical User Interface (GUI): A new GUI allows users to load data (including HDF5 containers from Oxford Instruments), preview images, reorder sequences, and adjust parameters without writing code.

3. Key Contributions

1. Software Re-engineering: Transitioning TrueEBSD from a standalone script to a fully integrated MTEX add-on, leveraging MTEX's robust data handling and crystallographic analysis tools.
2. Automation: Eliminating the need for manual control points, making the process fully automated and suitable for large datasets (e.g., 3D EBSD).
3. FAIR Principles: The code is maintained on GitHub, ensuring it is Findable, Accessible, Interoperable, and Reusable.
4. Enhanced Flexibility: The modular design allows users to mix and match different image types (e.g., combining FSE and BSE images acquired at different voltages or tilts) without modifying the core code.

4. Results: Case Studies

Case Study 1: WC-Co Composites (Phase Fraction & Contiguity)

• Objective: Accurately measure Cobalt (Co) phase fraction and Tungsten Carbide (WC) grain contiguity.
• Method: Aligned EBSD maps (20 kV) with FSE images (10 kV). The FSE images provided accurate phase contrast (Co vs. WC) which EBSD often misidentifies due to resolution limits.
• Findings:
  • Phase Fraction: Standard EBSD underestimated Co content by ~50% compared to FSE/Thermo-Calc models. The augmented map corrected this.
  • Contiguity: WC grain contiguity (the fraction of a grain's boundary shared with other WC grains) was recalculated.
    • Original EBSD overestimated contiguity in fine-grained microstructures (0.69 vs. 0.59 for BSE linear intercepts).
    • The FSE-augmented EBSD provided a more accurate value (0.49), revealing that EBSD misses thin Co layers, making WC grains appear to touch when they do not.
  • Conclusion: The method enables quantitative crystallographic measurements that are impossible with separate analysis.

Case Study 2: Copper Polycrystals (Grain Boundary Voids)

• Objective: Determine the susceptibility of specific grain boundaries (GBs) and triple points (TPs) to creep void formation.
• Method: Aligned EBSD maps with BSE images (0° tilt) where voids appear as dark pixels. Voids were segmented and mapped onto the EBSD orientation data.
• Findings:
  • Void Location: 90.2% of voids were located on or near grain boundaries.
  • Crystallographic Analysis: Misorientation analysis revealed that $\Sigma3$ twin boundaries (60°/<111>) were significantly underrepresented among void-containing boundaries, indicating they are highly resistant to void nucleation.
  • Triple Points: No specific triple junction type showed resistance or susceptibility; their distribution matched the background microstructure.
  • Significance: Demonstrated the ability to statistically correlate microstructural features with failure mechanisms using correlative data.

5. Significance

• Quantitative Accuracy: The tool resolves systematic errors in EBSD data (such as phase underestimation) by leveraging complementary imaging modalities, leading to more accurate microstructural metrics like grain contiguity.
• Automation & Scalability: By removing manual steps and optimizing code speed, TrueEBSD-MTEX enables the processing of large, complex datasets (e.g., 3D EBSD, in-situ experiments) that were previously too labor-intensive.
• Community Impact: As an open-source MTEX add-on, it lowers the barrier to entry for correlative microscopy, allowing researchers to perform advanced spatial registration and subsequent crystallographic analysis within a single, established ecosystem.
• Limitations Addressed: While the tool handles global distortions well, the authors note that highly localized, non-linear distortions (e.g., from surface topography on ion-milled samples) may still require manual intervention or are difficult to extrapolate to image edges.

In summary, TrueEBSD in MTEX represents a significant advancement in correlative microscopy, transforming image alignment from a manual, error-prone task into a robust, automated workflow that unlocks deeper quantitative insights into material microstructures.