Thanks Tim for helping out! I've ended up reiterating your solution to something similar. Just in case anyone would be interested, I will also post it here. (Ended up accessing an ndarray image by [Y, X] coordinates. Did use PixelLens Pycharm extension to debug how image arrays look like)

The code:

def find_scale_box_edges(image: np.ndarray):
    step = 5

    def find_last_black_pixel(image: np.ndarray, start, step):
        _x = start[0]
        _y = start[1]
        if (image[_y, _x][:3] != [0,0,0]).all():
            raise Exception

        while _x > 0 and _x < image.shape[1]:
            while _y > 0 and _y < image.shape[0]:
                if (image[_y + step[1], _x + step[0]][:3] != [0,0,0]).any():
                    return [_x, _y]
                else:
                    _x = _x + step[0]
                    _y = _y + step[1]

        raise Exception

    for x in range(image.shape[1]-step-1, step, -50):
        for y in range(image.shape[0]):
            if (image[y, x][:3] == [0, 0, 0]).any():

                # check N horizontal nearby pixels to see if its truly a line.
                upper_border_test = image[y:y+2, x-step:x+step][:,:,:3]
                if (upper_border_test[:, 0] == [[0, 0, 0], [255, 255, 255]]).all():
                    top_left_corner = find_last_black_pixel(image, [x,y], [-1, 0])
                    top_right = find_last_black_pixel(image, [x, y], [1, 0])
                    bottom_right_corner = find_last_black_pixel(image, top_right, [0, 1])

                    # scale_box_image = image[top_left_corner[1]:bottom_right_corner[1],
                    #                   top_left_corner[0]:bottom_right_corner[0]]

                    return [top_left_corner, bottom_right_corner]
    return [None, None]

Result: