opencvcorner-detection

How to detect only corners that has four quadrants

Opencv or any other image processing libs usually provide Harris corner detector, and it detects both corners types shown below:

enter image description here enter image description here

What would be the best way to only detect the first corner that has 4 quadrants and avoid the second type?

Solution

  • Is this useless?

    checkrer.png

    enter image description here

    import cv2
import numpy as np

filename = 'checker.png'
img = cv2.imread(filename)
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)


## Harris corner detector
gray_f = np.float32(gray)
dst = cv2.cornerHarris(gray_f,2,3,0.04)
dst = cv2.dilate(dst,None)
corner = dst>0.01*dst.max()

# Image for confirmation
# corner_img = img.copy()
# corner_img[corner] = [0,0,255]
# cv2.imwrite("corner.png", corner_img)


## low brightness area
ksize = (20, 20)
blue = cv2.blur(gray, ksize)
low = blue<150

# Image for confirmation
# low_img = img.copy()
# low_img[low] = [0,0,255]
# cv2.imwrite("low.png", low_img)


## cross point
cross_point = corner * low

# Image for confirmation
cross_img = img.copy()
cross_img[cross_point] = [0,0,255]
cv2.imwrite("cross.png", cross_img)

    cross.png

    enter image description here