Electrode Degradation — Traditional Image Processing Pipeline

This repository contains the baseline, rule-based pipeline used to assess electrode degradation from high-resolution microscope images. It sets a clear reference for performance before moving to data-driven methods. The implementation lives in the notebook Coating_defects_traditional.ipynb and uses OpenCV and scikit-image.

What the pipeline does

• Preprocess images to correct uneven illumination and low local contrast.

• Segment coating, substrate, and background into binary masks.

• Post-process masks with morphology and Boolean logic to derive defect classes.

• Detect delamination by intersecting dilated coating and substrate edges within background regions.

• Quantify defect lengths from contours and export results to CSV.

• Export visual overlays (PNGs) for each region.

Method (step-by-step)

1. Load image

Set the input path (e.g., 10-11318_6050-PG_17_0310.jpg) and the pixel size (default 0.73 μm/pixel).

2. Illumination correction (L channel)

Convert to LAB, take L, blur with a large Gaussian kernel (101×101), subtract, then normalize to [0, 255].

3. Contrast enhancement (CLAHE)

Apply CLAHE (clipLimit=3.0, tileGridSize=25×25) to the normalized L image.

4. Noise removal (before thresholding)

Use morphological opening with an elliptical kernel (5×5).

5. Adaptive Gaussian thresholding

Compute separate masks with tuned parameters:

• Coating:

binary (blockSize 11, C≈8)

• Substrate:

inverse binary (e.g., blockSize 11–15, C≈8–10)

• Background:

inverse binary (blockSize 11, C≈2)

6. Morphological cleanup

Apply opening to each thresholded mask; optionally filter by connected-component area.

7. Derived masks (Boolean logic)

• Covered coating:

From cleaned coating mask.

• Uncovered coating:

substrate_mask AND NOT coating_mask.

• Overall coating:

union of covered and uncovered.

8. Delamination detection

Dilate coating and substrate edges (3×3 rect kernel, iterations=2) and intersect them inside the background to get delaminated regions. Remove small components (e.g., area ≤220 px).

9. Length measurement (μm)

Trace contours (skimage.measure.find_contours), draw polylines, sum segment lengths, and convert to microns using pixel_size.

10. Outputs

• CSV: e.g., Stack_Frontview19_length_results.csv with region-wise lengths.

• PNG overlays per region (saved with region names).

Inputs and outputs

1. Input:

• A microscope image (e.g., .jpg/.png) placed in the repo or a /data folder.

2. Outputs:

• .png overlays for covered/overall coating and other regions.

• _length_results.csv with two columns: Region, Length (μm).

Requirements

• Python 3.9+

• OpenCV, NumPy, Matplotlib, Pandas, scikit-image