Applying Sobel filter from SciPy.ndimage

Question

I'm trying to apply the Sobel operator from SciPy.ndimage and replicate a result shown on Wikipedia, but the images are very different.

The result presented on Wikipedia shows the edges much more pronounced.

The code I'm using is listed below. Can this code be altered to agree with the result presented on Wikipedia? Original image as well as result image from Wikipedia are enclosed below.

import numpy as np
from PIL import Image
import matplotlib.pyplot as plt
from scipy.ndimage import filters


# Images from https://en.wikipedia.org/wiki/Sobel_operator
im_original = np.array(Image.open('Valve_original_(1).PNG').convert('L'))
im_sobel = np.array(Image.open('Valve_sobel_(3).PNG').convert('L'))

# Construct two ndarrays of same size as the input image
imx = np.zeros(im_original.shape)
imy = np.zeros(im_original.shape)

# Run the Sobel operator
# See https://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.sobel.html
filters.sobel(im_original,1,imx,cval=0.0)  # axis 1 is x
filters.sobel(im_original,0,imy, cval=0.0) # axis 0 is y

magnitude = np.sqrt(imx**2+imy**2)

# Construct the plot

fig = plt.figure(figsize=(10,8))

ax1 = fig.add_subplot(221)
ax1.set_title('Original (Wikipedia)')
ax1.axis('off')
ax1.imshow(im_original, cmap='gray')

ax2 = fig.add_subplot(222)
ax2.set_title('Sobel operator - as shown on Wikipedia')
ax2.axis('off')
ax2.imshow(im_sobel, cmap='gray')

ax3 = fig.add_subplot(224)
ax3.set_title('Sobel operator - from scipy.ndimage')
ax3.axis('off')
ax3.imshow(magnitude, cmap='gray')

plt.savefig('sobel.png')
plt.show()

Images

Original Image: Valve_original_(1).PNG

Result as presented on Wikipedia: Valve_sobel_(3).PNG

Answer 1

To put this question to rest I post an answer based on the comments above.

The Sobel filtered image of a steam engine as shown on Wikipedia has been processed in some additional manner not specified, and can therefore not be fully reproduced. Most likely the original RGB image was first converted to a grayscale and then clamped.

Looking at the intensity histogram for the Sobel filtered image obtained from SciPy.ndiamage, see Figure below, most pixels are centred around an intensity of 3.5. Applying a clamping value of 50 produce an image closer to what is shown on the Wikipedia page.