I am working on a task where I need to detect the four corners of each square in an image and highlight each corner as a single pixel.
Each filter is intended to respond to a specific corner orientation. Applying the filters produces four response maps, which have to be combined into a single 4-channel output array of shape (H, W, 4), ordered as Z, D, L, S (top, right, left, bottom).
Current implementation successfully generates response maps, but I am unsure how to convert these responses into exactly one pixel per corner.
This is the image I am using for testing:
And this is my output.
def corners_of_square(img: np.ndarray) -> np.ndarray: # doesn't work
img_cp = img.copy()
if img_cp.ndim == 3:
img_cp = cv.cvtColor(img_cp, cv.COLOR_BGR2GRAY).astype(np.float32)
img_cp = img_cp.astype(np.float32)
Z = np.array([ # top
[ 1, 1, 1],
[ 1, -1, -1],
[ 1, -1, -1]
], dtype=np.float32)
D = np.array([ # right
[ 1, 1, 1],
[-1, -1, 1],
[-1, -1, 1]
], dtype=np.float32)
L = np.array([ # left
[ 1, -1, -1],
[ 1, -1, -1],
[ 1, 1, 1]
], dtype=np.float32)
S = np.array([ # bottom
[-1, -1, 1],
[-1, -1, 1],
[ 1, 1, 1]
], dtype=np.float32)
L = np.array([[1,-1,-1],[1,-1,-1],[1,1,1]], dtype=np.float32) # L = left
S = np.array([[-1,-1,1],[-1,-1,1],[1,1,1]], dtype=np.float32) # S = bottom
rZ = cv.filter2D(img_cp , cv.CV_32F, Z)
rD = cv.filter2D(img_cp , cv.CV_32F, D)
rL = cv.filter2D(img_cp , cv.CV_32F, L)
rS = cv.filter2D(img_cp , cv.CV_32F, S)
out = np.stack([rZ, rD, rL, rS], axis=2)
#out[i, j] = [Z_response, D_response, L_response, S_response]
return out
if __name__ == '__main__':
rotated_squares_img = cv.imread(str(UTILS/"rotirani_kvadrati.png")).astype(np.float32) / 255
corners_img = corners_of_square(rotated_squares_img)
cv.imshow("Rotated squares 1", rotated_squares_img)
cv.imshow("Rotated squares 2", corners_img)
cv.waitKey(0)
cv.destroyAllWindows()
