Python PIL ValueError：图像不匹配

Question

I'm playing around with PIL and came across this problem and I can't see where in the docs I'm going wrong. Here is my simple code

from PIL import Image
from PIL.ImageChops import difference

imageA = Image.open("image1.png")
imageB = Image.open("image2.png")

if imageA.size == imageB.size:
    diff = difference(imageA, imageB)
    diff.save("test.png")

which gives me the error

Traceback (most recent call last):
  File "C:\[XXX]\box-test.py", line 8, in <module>
    diff = difference(imageA, imageB)
  File "C:\Python32\lib\site-packages\PIL\ImageChops.py", line 123, in difference
    return image1._new(image1.im.chop_difference(image2.im))
ValueError: images do not match

Any help would be appreciated

Answer 1

The documentation for this function doesn't tell much in fact. So let me try to clarify it a little. First, the sizes of the images are irrelevant to whether the function works or not, it internally checks for a size that both images fit.

Now, when can you actually compare the images by using the function ImageChops.difference ?

First, both images have to have pixels that can be stored in an unsigned byte. This is a very common type of image, but this excludes comparison between images even if they are the same mode. So, you cannot compare an image x and y when one or /both/ of them have a mode of: F , I , I;16 , I;16L , I;16B , BGR;15 , BGR;16 , BGR;24 , or BGR;32 . Just to make it clear: it doesn't matter if both images are in the same mode if they happen to be in one of the modes above, the function will refuse to work.

So, the comparison can be done when the images are in the modes 1 , P , L , LA , RGB , RGBA , RGBX , RGBa , CMYK , or YCbCr as long as they have the same number of bands. This means the images don't have to have the same mode to be compared. For instance, difference(x.convert('CMYK'), x.convert('RGBA')) or difference(x.convert('1'), x.convert('P')) work just fine. Of course this means difference(x.convert('LA'), x.convert('L')) , fails. Finally, the resulting image will always have the mode equal to the first image passed to the function.

This is valid at least for the PIL 1.1.7.

Answer 2

只需验证每个图层和图像的每个图像的大小（例如 1024 x 1024）