OpenCV pointPolygonTest() not working correctly

I want to split an image in 6 equal areas (as seen in the picture) and then I want to know to which area a point belongs to. I'm using pointPolygonTest() to check whether a point is inside a contour or not, the thing is for certain areas it works but for some others it doesn't.

Below I show my code to do so. I have included a test where it works correctly and other where it doesn't because it says the point does not belong to any area, while in the picture it can clearly be seen it does.

Input:

img_path = "../tokyo.jpg"
im = cv2.imread(img_path)
im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB)

xs = [0,im.shape[0]//2,im.shape[0]]
ys = [0,im.shape[1]//3,im.shape[1]//3*2,im.shape[1]]

contours = []

for i in range(len(xs)-1):
    for j in range(len(ys)-1):
        box = np.array([[xs[i],ys[j]],[xs[i],ys[j+1]],[xs[i+1],ys[j]],[xs[i+1],ys[j+1]]],dtype=np.int32)
        contours.append(box)

for c in contours:
    im = cv2.rectangle(im,(c[0]),(c[3]), (0, 255, 0), 10) # green
# im = cv2.circle(im, (400,400), radius=1, color=(255, 0, 0), thickness=10)
plt.imshow(im)
plt.show()

Output:

Input:

for i,c in enumerate(contours):
    distance = cv2.pointPolygonTest(c, (400,400), measureDist=False)
    if distance >= 0:
        print(f"Point is inside the contour {i+1}")
    else:
        print(f"Point is outside the contour {i+1}")

Output:

Point is outside the contour 1
Point is outside the contour 2
Point is outside the contour 3
Point is outside the contour 4
Point is outside the contour 5
Point is outside the contour 6

From the picture it is clear it should say "Point is inside the contour 5". Below you can find a code where it does work and just by changing the distance a few pixels, so I don't know why does this happen. Maybe I'm missing something. Any help would be appreciated.

Input: changed the point from (400,400) to (300,400)

for i,c in enumerate(contours):
    distance = cv2.pointPolygonTest(c, (300,400), measureDist=False)
    ...

Output:

Point is outside the contour 1
Point is inside the contour 2
Point is outside the contour 3
Point is outside the contour 4
Point is outside the contour 5
Point is outside the contour 6

Update: I just tested a few more times and realized it does work when the point is in the left-half of the image, but when is on the right it doesn't. No idea for the reason behind this.

Your code:

box = np.array([
    [ xs[i],   ys[j]   ],
    [ xs[i],   ys[j+1] ],
    [ xs[i+1], ys[j]   ],
    [ xs[i+1], ys[j+1] ]
], dtype=np.int32)

This defines the points in the following order (assuming you've fixed the width/height mixup):

• top left
• top right
• bottom left
• bottom right

That's not a contour, that's a self-intersecting hourglass.

The points need to be ordered correctly. Just reorder the last two coordinates.

Note: no, OpenCV contours are nothing special. they're just numpy arrays. They should have the shape (N, 1, 2) but (N, 2) will probably work too.

im.shape[0] - is height of the image
im.shape[1] - width

xs = [0,im.shape[1]//2,im.shape[1]]
ys = [0,im.shape[0]//3,im.shape[0]//3*2,im.shape[0]]