DOC: clarify that prewitt filter does not return the magnitude by default

[dschmitz89]

Issue with current documentation:

Like the sobel filter (#18444), prewitt by default computes a one dimensional transform. We should show how to compute the magnitude as that is typically the more useful image (cf. #18444).

from scipy import ndimage, datasets
import matplotlib.pyplot as plt
from skimage import filters

def prewitt_2d(image):
    image = image.astype('int32')
    dx = ndimage.prewitt(image, 0)  # horizontal derivative
    dy = ndimage.prewitt(image, 1)  # vertical derivative
    mag = np.hypot(dx, dy)  # magnitude
    mag *= 255.0 / np.max(mag)
    return mag

fig = plt.figure(figsize=(12, 3))
plt.gray()  # show the filtered result in grayscale
ax1 = fig.add_subplot(141)
ax2 = fig.add_subplot(142)
ax3 = fig.add_subplot(143)
ax4 = fig.add_subplot(144)


ascent = datasets.ascent()
ax1.imshow(ascent)
ax2.imshow(ndimage.prewitt(ascent))
ax2.set_title("scipy prewitt default (1D)")
ax3.imshow(filters.prewitt(ascent))
ax3.set_title("skimage prewitt default")
ax4.imshow(prewitt_2d(ascent))
ax4.set_title("scipy prewitt 2D magnitude")
for ax in [ax1, ax2, ax3, ax4]:
    ax.set_axis_off()
plt.show()

[Ashad001]

Hi, can you emphasize what has to be done, the description seems to be a bit vague. Thankyou!

[dschmitz89]

The root issue is that scipy.ndimage.prewitt by default computes a one dimensional Prewitt filter. This is not the "typical" use case in image processing. For example, skimage implements the magnitude instead (see images above).

In #18444, the same issue was fixed for the sobel filter. The documentation/examples should emphasize how to get the "typical" Prewitt filter from the function.

What needs to be done is a documentation PR. The Sobel filter PR can serve as an example. For general SciPy development, I recommend having a look at the quickstart and documentation related parts of the developer docs.