Change a cell in array into the mean of it's surroundings, for an entire array

Question

I want to go through an array and find anomalies (specifically, values higher than 100).

I then want the anomalous cell to be replaced by the mean value of the surrounding cells.

Let's say the input is:

[6, 28, 33]
[20, 100, 41]
[87, 3, 39]

I want to change that 100 cell into the integer mean of

6, 28, 33, 20, 41, 87, 3, 39

which is 32

The output should be:

[6, 28, 33]
[20, 32, 41]
[87, 3, 39]

If it matters, my real array is 256*256 cells, and has ~500 values I want to change

Last point!

I know that there would be a problem fixing up the edges, so I don't care whether they are thrown out completely or some clever piece of code can average without them.

Here is my attempt:

import numpy as np
array = np.random.randint(100, size=(256, 256))
    for x in array
    if x>=100:
        x = np.mean(x+1,x-1) 
#This is where I got stuck... trying to define the surrounding cells

Answer 1

So I know this may not be the most efficient solution, but it works. If the anomaly is on the edge of the array, it just gets the mean of the array before or after it, not both.

inp =   [[6, 10, 33],
        [20, 100, 41],
        [87, 3, 39]]


for i in range(len(inp)):
    for j in range(len(inp[i])):
        if inp[i][j] >= 100:
            if i == len(inp) - 1:
                inp[i][j] = int((sum(inp[i])-inp[i][j] + sum(inp[i-1])) / (len(inp[i])-1 + len(inp[i-1])))
            elif i == 0:
                inp[i][j] = int((sum(inp[i])-inp[i][j] + sum(inp[i+1])) / (len(inp[i])-1 + len(inp[i+1])))
                print(inp[i][j])
            else:
                inp[i][j] = int((sum(inp[i])-inp[i][j] + sum(inp[i+1]) + sum(inp[i-1])) / (len(inp[i])-1 + len(inp[i+1]) + len(inp[i+1])))

Again, this solution isn't particularly efficient or pretty, but it does the job.