Image derivative with smoothed derivative kernel

smooth_derivative.py

#!/usr/bin/env python3
# -*- coding: utf8 -*-
#https://stackoverflow.com/questions/29731726/how-to-calculate-a-gaussian-kernel-matrix-efficiently-in-numpy
import numpy as np
from scipy.ndimage import convolve

def gauss_kernel(size=5, sigma=None):
    if sigma is None:
        sigma = size/5
    xs = np.arange(-size//2 + 1, size//2 + 1)
    grid_x, grid_y = np.meshgrid(xs, xs)
    kernel = np.exp(-(grid_x**2 + grid_y**2)/(2 * sigma**2))
    return kernel/kernel.sum()

def smooth_derive(img, along_x = False, size=5, sigma=None):
    kernel = gauss_kernel(size, sigma)
    w = np.array([[-1, 0, 1]])/2
    if not along_x:
        w = w.T
    d_op = convolve(kernel, w) # construct smoothed derivative operator
    deriv_img = convolve(img, d_op) # convolve the derivative operator
    return deriv_img

if __name__ == "__main__":
    import argparse, sys
    import matplotlib.pyplot as plt
    import cv2

    if sys.version_info.major < 3:
        input = raw_input

    parser = argparse.ArgumentParser()
    parser.add_argument("src_img", metavar="IMAGE_PATH", help="source image path")
    parser.add_argument("-x", "--along-x", dest="along_x", action="store_true", help="along x direction")
    args = vars(parser.parse_args())

    img = cv2.imread(args["src_img"], cv2.IMREAD_GRAYSCALE)
    derv = smooth_derive(img, args["along_x"], 15)
    plt.interactive(True)
    plt.imshow(derv, cmap='gray')
    input("press enter/return to exit")