[Python] Contour completion for complex contour

I have a discontinuous circular contour, and I want to calculate the area enclosed by this contour. I applied OpenCV's contourArea to compute the area, but due to the discontinuity in the contour, errors occur during the calculation. To address this, I tried completing the contour using neighboring point interpolation (Delaunay triangulation). However, this method causes distortion in the contour. How can I effectively complete my contour?

This is the code I used for calculating the area

import cv2
import numpy as np
import matplotlib.pyplot as plt


image_path = "processed_images2/cropped_good4.jpg_640x640.jpg"
image = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)


_, binary = cv2.threshold(image, 127, 255, cv2.THRESH_BINARY)


contours, _ = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)


total_area = sum(cv2.contourArea(cnt) for cnt in contours)


image_contours = cv2.cvtColor(image, cv2.COLOR_GRAY2BGR)


cv2.drawContours(image_contours, contours, -1, (0, 255, 0), thickness=2)


plt.figure(figsize=(6,6))
plt.imshow(image_contours)
plt.title(f"Estimated Area: {total_area:.2f} pixels²")
plt.axis("off")
plt.show()


here's the code that I implement the Delaunay triangulation for contour

import scipy.spatial
import cv2

image = cv2.imread("processed_images2\\cropped_147.jpg_640x640.jpg", cv2.IMREAD_GRAYSCALE)
edges = cv2.Canny(image, 50, 150)


y, x = np.where(edges > 0)
contour_points = np.column_stack((x, y))

fixed_image = image.copy()
tri = scipy.spatial.Delaunay(contour_points)
for simplex in tri.simplices:
pts = contour_points[simplex]
cv2.polylines(fixed_image, [pts], isClosed=True, color=255, thickness=1)

plt.figure(figsize=(6, 6))
plt.imshow( fixed_image, cmap='gray')
# plt.title(f"OR Operation: {image_name}")
plt.axis("off")
plt.show()


This is the input image



output



At the same time, I also tested KNN and KD-tree methods to complete the contour, but the results were not ideal. I hope the final contour is a closed curve without distortion.

KD tree:

import cv2
import numpy as np
from scipy.spatial import KDTree


image = cv2.imread("processed_images2\\cropped_147.jpg_640x640.jpg", cv2.IMREAD_GRAYSCALE)
edges = cv2.Canny(image, 50, 150)


y, x = np.where(edges > 0)
contour_points = np.column_stack((x, y))


tree = KDTree(contour_points)


missing_points = np.array([[100, 150], [120, 200]]) # (x, y)


_, nearest_idx = tree.query(missing_points)


fixed_image = image.copy()
for i, idx in enumerate(nearest_idx):
x_nearest, y_nearest = contour_points[idx]
cv2.circle(fixed_image, (missing_points[0], missing_points[1]), 2, 255, -1)

plt.figure(figsize=(6, 6))
plt.imshow( fixed_image, cmap='gray')
# plt.title(f"OR Operation: {image_name}")
plt.axis("off")
plt.show()


KNN

from sklearn.neighbors import KNeighborsClassifier


knn = KNeighborsClassifier(n_neighbors=3)
knn.fit(contour_points, np.ones(len(contour_points)))


predictions = knn.predict(missing_points)


for i, pred in enumerate(predictions):
if pred == 1:
cv2.circle(fixed_image, (missing_points[0], missing_points[1]), 2, 255, -1)

plt.figure(figsize=(6, 6))
plt.imshow( fixed_image, cmap='gray')
# plt.title(f"OR Operation: {image_name}")
plt.axis("off")
plt.show()

Continue reading...