使用PIL保存损坏的图像

我遇到一个问题，即操纵图像像素会导致图像被损坏...

I am having an issue whereby manipulating image pixels is causing a corrupted image to be saved...

所以，我用PIL打开图像，然后转换它到一个NumPy数组：

So, I open an image using PIL, and then convert it to a NumPy array:

image = Image.open("myimage.png")
np_image = np.asarray(image)

然后，我转置图片，从转换它[x] [y] [频道] 到 [频道] [x] [y] ：

Then, I transpose the image, to convert it from [x][y][channel] to [channel][x][y]:

pixels = np.transpose(np_image, (2, 0, 1))

如果我然后将此图像转换回 [x] [y] [channel] ，请从此数组创建PIL图像，然后保存图片：

If I then transpose this image back to [x][y][channel], create a PIL image from this array, and then save the image:

image1 = np.transpose(pixels, (1, 2, 0))
image2 = Image.fromarray(image1, 'RGB')
image2.save('image2.png')

然后保存的图像与myimage.png相同。

Then the image saved is identical to "myimage.png".

但是，如果不是上面的代码，我首先指定像素到el在一系列图像中：

However, if instead of the above code, I first assign pixels to an element in an array of images:

images = np.zeros([10, 3, 50, 50]) # The images are 50x50 with 3 channels
images[0] = pixels
image3 = np.transpose(images[0], (1, 2, 0))
image4 = Image.fromarray(image3, 'RGB')
image4.save('image4.png')

然后image4 .png已损坏。它显示如下：

Then "image4.png" is corrupted. It appears as follows:

而myimage .png实际上是：

Whereas "myimage.png" is actually:

那么为什么如果我直接转置像素时保存图像，保存的图像是预期的，但是当我设置像素$ c $时c>到数组 images 中的第一个元素，然后转置此图像，保存的图像已损坏？

So why is it that if I save the image when I directly transpose pixels, the image saved is as expected, but when I set pixels to the first element in the array images, and then transpose this image, the saved image is corrupted?

谢谢！