4

I have this image:

src

What I am trying to do is to detect the center of mass of the inner contour (number 3) inside it.

This is the code I have right now:

import cv2
import numpy as np

im = cv2.imread("three.png")

imgray = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(imgray, 127, 255, 0, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)
_, contours, hierarchy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_NONE)

cnts = cv2.drawContours(im, contours[1], -1, (0, 255, 0), 1)

cv2.imshow('number_cnts', cnts)
cv2.imwrite('number_cnts.png', cnts)

m = cv2.moments(cnts[0])
cx = int(m["m10"] / m["m00"])
cy = int(m["m01"] / m["m00"])

cv2.circle(im, (cx, cy), 1, (0, 0, 255), 3)

cv2.imshow('center_of_mass', im)
cv2.waitKey(0)
cv2.imwrite('center_of_mass.png', cnts)

This is the (wrong..) result:

result

Why the center of mass has been draw in the left part of the image instead of in the (more or less) center ?

Any solution to this ?

Jeru Luke
  • 20,118
  • 13
  • 80
  • 87
lucians
  • 2,239
  • 5
  • 36
  • 64

1 Answers1

5

You can try by taking the average of contour points, mentioned here.

  imgray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
  ret, thresh = cv2.threshold(imgray, 127, 255, 0, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)
  _, contours, hierarchy = cv2.findContours(thresh, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)

  cnts = cv2.drawContours(image, contours[0], -1, (0, 255, 0), 1)

  kpCnt = len(contours[0])

  x = 0
  y = 0

  for kp in contours[0]:
    x = x+kp[0][0]
    y = y+kp[0][1]

  cv2.circle(image, (np.uint8(np.ceil(x/kpCnt)), np.uint8(np.ceil(y/kpCnt))), 1, (0, 0, 255), 3)


  cv2.namedWindow("Result", cv2.WINDOW_NORMAL)
  cv2.imshow("Result", cnts)
  cv2.waitKey(0)
  cv2.destroyAllWindows()

result

zindarod
  • 6,328
  • 3
  • 30
  • 58
  • This works for this case because your shape is fairly rounded so the contour points are pretty nicely distributed. If you had a 'D' for example, there'd be no contour points on the left line of the 'D', so the center of mass would be hugely biased. – Denis Sep 26 '19 at 21:26
  • a simpler way of getting the COM is by using numpy's `mean`: `x, y = contour.mean(axis=0)[0]` – Griffin Sep 03 '21 at 10:52