numpy 数组的平均像素减法

Question

I have a numpy array of grayscale images. The shape of the array is

imgs.shape
(100, 1, 300, 300)

which represents 100 images of grayscale (1 channel) and size 300x300.

I want to do mean pixel subtraction for this array. That means I want to substract the mean pixel from all images.

I want the mean over all images.

For a single image that would be something like this:

X_mean = X.mean(0)
X -= X_mean

How can I do that for my array?

Answer 1

If its mean of one image at a time, we could compute the mean across the last two axes keeping the dimensions (that eases up code for subtraction later on) and simply subtract from original input, like so -

imgs -= imgs.mean(axis=(-2,-1),keepdims=1)

If its mean across all images, include the first axis too -

imgs -= imgs.mean(axis=(0,-2,-1),keepdims=1)

Note that the mean values would be by default as float values. So, if imgs is not of float type, we need to make a copy of imgs as float dtype and then subtract from it, or convert the mean values to be of same dtype as imgs and then subtract.

Sample run -

In [188]: imgs = np.random.randint(0, 255,(10,1,30,30))

In [189]: out = imgs - imgs.mean(axis=(0,-2,-1),keepdims=1)

In [190]: out.shape
Out[190]: (10, 1, 30, 30)

In [191]: out.dtype
Out[191]: dtype('float64')