Convolve to find axis aligned corners. Also, find nearest pixel

corner_nearest_neighbor.py

   def convert_bin_img_to_corners(img):
        '''
        Mark corners with bias being pixels origin as (x,y): (0.5, 0.5)
        :param img:     Binary image of field
        :return:        Image of corners marked with 255, all else is 0
        '''
        kernel = np.ones(
            (2, 2), np.float32)  # Make convolution kernel 2x2 of 0.25s
        ret, img_thresh = cv2.threshold(
            img, 1, 1, cv2.THRESH_BINARY)  # Make img into [0,1]
        dst = cv2.filter2D(
            img_thresh, -1, kernel,
            anchor=(0, 0))  # convolving marks corners with a 1 or a 3
        idx = (dst == 2) + (dst == 4) # Get rid of 2s and 4s in the dst image
        dst[idx] = 0
        idx2 = (dst > 0)  # set dst to [0,255]
        dst[idx2] = 255
        nonzero = cv2.findNonZero(dst)
        return dst

    def find_nearest_white(img, xy_point):
        '''
        Given an image (usually [0,1] or [0,255]) find the nearest non-zero pixel to xy_point
        :param img:         Image, usually binarized to [0,1] or [0,255]
        :param xy_point:    Tuple of (x,y)
        :return:            Index (x,y) tuple of nearest non-zero point.
        '''
        nonzero = cv2.findNonZero(img)
        distances = np.sqrt((nonzero[:, :, 0] - xy_point[0]) ** 2 + (nonzero[:, :, 1] - xy_point[1]) ** 2)
        nearest_index = np.argmin(distances)
        return nonzero[nearest_index][0] + 0.5