使用scikit-image模块时遇DLL加载失败错误求助

Hey there, let's get this DLL load issue sorted out for you. The error trace points directly to a problem with the PIL (Pillow) library's _imaging module failing to load—this is super common with older Python versions like 3.6.4 where package compatibility can get tricky. Here are step-by-step fixes to try:

1. Reinstall a Pillow Version Compatible with Python 3.6.4 #

Older Python versions need matching older Pillow releases. Let's swap out your current Pillow for one that plays nice:

• First, open your Anaconda Prompt, activate your working environment (if you're not using base), then uninstall the current Pillow:

  conda uninstall pillow -y
  

• Install a verified compatible version (Pillow 5.4.1 works great with Python 3.6 on Windows):

  conda install pillow=5.4.1 -y
  

  If conda gives you trouble with this version, try using pip instead:

  pip install pillow==5.4.1
  

2. Ensure Your Environment Has numpy-mkl (Not Regular numpy) #

You mentioned trying numpy-mkl, but let's double-check it's properly set up:

• List your current numpy package to confirm:

  conda list numpy
  

• If you see just numpy (not numpy-mkl), uninstall the regular version and install mkl-enabled numpy:

  conda uninstall numpy -y
  conda install numpy-mkl -y
  

3. Install the Correct Visual C++ Redistributable #

Python 3.6 is compiled with Visual Studio 2015, so you need the matching redistributable package (even if you installed newer VC libraries):

• Make sure you have the 64-bit Visual C++ Redistributable for Visual Studio 2015 installed—this is required for the DLLs used by Pillow and scikit-image to load correctly. If you previously installed newer versions (like 2017/2019), this older version is still needed alongside them.

4. Test the Fix with a Minimal Code Snippet #

After making the above changes, test if PIL loads first (since that's the root issue):
Open Spyder and run this minimal code:

from PIL import Image
# Try opening your image to verify PIL works
test_img = Image.open('mire.jpg')
test_img.show()

# Now test scikit-image
from skimage import io
img = io.imread('mire.jpg', as_grey=True)
io.imshow(img)

If PIL loads and shows the image, scikit-image should work without the DLL error.

5. Create a Clean Environment (If All Else Fails) #

If your current environment has conflicting packages, a fresh environment often fixes things:

• Create a new environment with Python 3.6.4:

  conda create -n skimage_test python=3.6.4 -y
  

• Activate it and install the required packages:

  conda activate skimage_test
  conda install scikit-image numpy-mkl pillow=5.4.1 -y
  

• Switch Spyder to use this new environment (go to Tools > Preferences > Python interpreter and select skimage_test), then run your code again.

内容的提问来源于stack exchange，提问作者nizar aoufi