Optimizing ijGranulometry Parameters for Reliable Results

Troubleshooting Common Issues in ijGranulometry Analyses

1. Poor segmentation / unclear particle boundaries

• Cause: Low contrast, uneven illumination, or inappropriate thresholding.
• Fixes:
  • Apply background correction (e.g., “Subtract Background” or flat-field correction).
  • Use contrast enhancement (Brightness/Contrast or CLAHE).
  • Try different threshold methods (Otsu, Yen, Triangle) and use manual thresholding if needed.
  • Smooth lightly (Gaussian blur, radius 1–2 px) before thresholding to reduce noise.

2. Over-segmentation (particles split)

• Cause: Noise, texture inside particles, or too aggressive watershed/separation.
• Fixes:
  • Pre-filter using median or Gaussian filters to remove small texture.
  • Use morphological closing to fill small holes before segmentation.
  • Adjust watershed markers: erosion to separate touching particles less aggressively; use distance transform with an appropriate peak suppression.
  • Merge small fragments post-hoc by size filtering or connectivity rules.

3. Under-segmentation (particles merged)

• Cause: Insufficient edge detection or low contrast between touching particles.
• Fixes:
  • Increase contrast between particles and background (local thresholding, CLAHE).
  • Apply edge-enhancing filters (Sobel, Laplacian) before separation.
  • Use watershed with better marker control (find local maxima on distance map).
  • Manually split with the ROI tools for critical areas.

4. Incorrect particle size distribution (bias toward small or large sizes)

• Cause: Wrong calibration, scale not set, size filtering excluding valid particles, or artifacts counted as particles.
• Fixes:
  • Verify image calibration (Analyze > Set Scale) using known scale bars.
  • Remove artifacts using size and circularity filters; inspect excluded objects.
  • Ensure thresholding doesn’t fragment particles into many tiny objects.
  • Run granulometry on multiple representative images and report variability.

5. High noise / many tiny spurious particles

• Cause: Sensor noise, dust, or overly sensitive detection.
• Fixes:
  • Apply noise-reduction filters (median, minimum/maximum).
  • Remove objects below a realistic size cutoff (based on calibration).
  • Clean images physically (better sample prep) and use dark-frame subtraction if necessary.

6. Irregular particle shapes causing measurement errors

• Cause: Non-spherical grains produce misleading single-parameter metrics.
• Fixes:
  • Report multiple shape descriptors (area, Feret diameter, aspect ratio, circularity).
  • Use granulometric curve methods (morphological opening sizes) rather than solely particle counts.
  • Consider oriented bounding boxes or convex hull metrics for elongated particles.

7. Reproducibility issues between runs or users

• Cause: Different preprocessing, thresholds, or parameter choices.
• Fixes:
  • Script the workflow in ImageJ macros or use the ijGranulometry plugin scripting to fix parameters.
  • Save and share threshold presets and macro files.
  • Use consistent calibration and capture settings.

8. Performance and memory problems with large images

• Cause: Very large TIFFs or many images processed at once.
• Fixes:
  • Process images in tiles or downsample for preliminary checks.
  • Increase ImageJ memory (Edit > Options > Memory & Threads).
  • Use 64-bit Fiji distribution and close unnecessary windows/plugins.

9. Unexpected plugin errors or crashes

• Cause: Version incompatibility or corrupted installation.
• Fixes:
  • Use the latest Fiji distribution (bundled ImageJ) and update plugins via Help > Update.
  • Reinstall ijGranulometry plugin from a trusted source.
  • Check Fiji’s log window for stack traces and search for known issues online.

10. Interpreting granulometry output incorrectly

• Cause: Misunderstanding metrics (e.g., D10, D50, D90) or cumulative vs. differential plots.
• Fixes:
  • Report and explain which metrics you use (median, mean, percentiles).
  • Plot both cumulative and differential distributions.
  • Normalize counts to area or volume if comparing different magnifications.

If you want, I can convert these fixes into an ImageJ/Fiji macro that implements common preprocessing and thresholding steps for ijGranulometry.