Best way to detect if checkbox is ticked

Question

My work:

1. Scan the paper
2. Check horizontal and vertical line
3. Detect checkbox
4. How to know checkbox is ticked or not

At this point, I thought I could find it by using Hierarchical and Contours: Below is my work

for i in range (len( contours_region)):     #I already have X,Y,W,H of the checkbox through 
    #print(i)                               #cv2.connectedComponentsWithStats
    x = contours_region[i][0][1]            #when detecting checkbox
    x_1 = contours_region[i][2][1]
    y = contours_region[i][0][0]
    y_1 = contours_region[i][2][0]

    image_copy= image.copy()

    X,Y,W,H = contours_info[i]
    cv2.drawContours(image_copy, [numpy.array([[[X,Y]],[[X+W,Y]],[[X+W,Y+H]],[[X,Y+H]]])], 0, (0,0,255),2)
    gray = cv2.cvtColor(image_copy, cv2.COLOR_BGR2GRAY)
    ret,bw = cv2.threshold(gray,220,255,cv2.THRESH_BINARY_INV)
    
    contours,hierarchy = cv2.findContours(bw[x:x_1, y:y_1], cv2.RETR_CCOMP,1)
    
    print('-----Hierarchy-----')
    print(hierarchy)
    print('-----Number of Contours : '+ str(len(contours)))
    cv2.imshow('a', image_copy)
    cv2.waitKey(0)

I got this result (some high contours, some high hierarchy)

-----Hierarchy-----
[[[-1 -1  1 -1]
  [ 2 -1 -1  0]
  [ 3  1 -1  0]
  [ 4  2 -1  0]
  [ 5  3 -1  0]
  [ 6  4 -1  0]
  [ 7  5 -1  0]
  [-1  6 -1  0]]]
-----Number of Contours : 8

---

Another result:

Low Contours, Low Hierarchy

-----Hierarchy-----
[[[-1 -1  1 -1]
  [ 2 -1 -1  0]
  [-1  1 -1  0]]]
-----Number of Contours : 3

---

However, it's not perfect some case where it's not ticked but still got a really high result

[[[-1 -1  1 -1]
  [ 2 -1 -1  0]
  [ 3  1 -1  0]
  [ 4  2 -1  0]
  [ 5  3 -1  0]
  [-1  4 -1  0]]]
-----Number of Contours : 6

---

In general, After review the whole data, the gap is not convincing between ticked and not ticked. Around 30% of boxes, giving the wrong result. Therefore, really wish to have a better method.

Answer 1

I think erode function help you. Use erosion to make the ticks bigger then count the non zero pixels. Here You can find the basics:

import cv2 
import numpy as np
from google.colab.patches import cv2_imshow
img = cv2.imread("image.png");
cv2_imshow(img)
kernel = np.ones((3, 3), np.uint8) 

better_image = cv2.erode(img,kernel)
cv2_imshow(better_image)