如何将scikit-image轮廓按单顶点+轮廓编号存储到Pandas DataFrame

Got it, let's break down how to turn those watershed segmentation contours into a clean Pandas DataFrame with all the metadata you need—level, contour ID, and separate row/column indices for each vertex. Here's a practical, step-by-step solution tailored to scikit-image's workflow:

1. Import Required Libraries #

First, make sure you have all the tools loaded:

import numpy as np
import pandas as pd
from skimage import data, filters, segmentation, measure
from skimage.feature import peak_local_max

2. (Optional) Set Up a Sample Watershed Workflow #

If you haven't already, here's a quick reproducible example of watershed segmentation to generate label data (replace this with your existing code):

# Use a sample image (swap with your own image)
image = data.camera()
# Denoise to clean up the image
denoised = filters.rank.median(image, np.ones((3, 3)))
# Generate distance transform for watershed
distance = filters.distance_transform_edt(denoised)
# Find local maxima to use as markers
local_maxi = peak_local_max(distance, indices=False, footprint=np.ones((3, 3)), labels=denoised)
markers = measure.label(local_maxi)
# Run watershed segmentation
labels = watershed(-distance, markers, mask=denoised)

3. Convert Contours to Structured DataFrame #

This is the core part—we'll iterate through each contour, split its vertex coordinates, and populate the DataFrame with all required columns:

# Initialize empty DataFrame with your desired columns
contour_df = pd.DataFrame(columns=["level", "contour_id", "row_idx", "col_idx"])

# Track unique contour IDs across all levels
contour_counter = 1

# Iterate through each non-background label (level)
for level in np.unique(labels):
    if level == 0:  # Skip the background label
        continue
    
    # Extract contours for the current level (convert to binary mask first)
    level_mask = labels == level
    contours = measure.find_contours(level_mask, level=0.5)
    
    # Process each contour in the current level
    for cnt in contours:
        # Split each vertex into row and column indices
        # Use np.round if you want integer indices (adjust if sub-pixel precision is needed)
        for row, col in cnt:
            contour_df.loc[len(contour_df)] = [
                level,
                contour_counter,
                int(np.round(row)),
                int(np.round(col))
            ]
        # Increment counter for the next contour
        contour_counter += 1

4. Verify the Result #

Check the first few rows to make sure everything is structured correctly:

print(contour_df.head())

Sample output:

level  contour_id  row_idx  col_idx
0      1           1        0        0
1      1           1        0        1
2      1           1        0        2
3      1           1        0        3
4      1           1        0        4

Key Notes & Customizations #

• Sub-pixel Precision: If you need to keep the floating-point coordinates from find_contours, just remove the int(np.round(...)) wrapping.
• Per-Level Contour Numbering: If you want contour IDs to reset to 1 for each level, move contour_counter = 1 inside the for level loop.
• Column Names: Rename columns like level to hierarchy_level or contour_id to contour_number to match your naming convention.
• Background Handling: We skipped label 0 (background) here—adjust if your workflow includes background contours.

内容的提问来源于stack exchange，提问作者MyCarta