Subsetting multiple matrices of a specified size

Consider my 20x20 matrix:

M1 <- matrix(rnorm(400), nrow=20, ncol=20)

What I am trying to do is:

Subset a 5x5 block of this matrix
Calculate the median of this 5x5 block
Replace the middle value of this 5x5 block with that median value
Rinse and repeat with every 5x5 block in the matrix

So far I have managed to do this for all the columns with a fixed block of rows:

x <- 0:20

for (i in seq_along(x)) {
  matrixtest[i] <- median(as.matrix(M1[1:5, (1 + i):(5 + i)]))
  M1[3, 3 + seq_along(x)] <- matrixtest[seq_along(x)]
}

However, the problem is I want to also do this going down each row, and I can't figure out a way to do it without manually specifying the rows as I have done below.

matrixtest2[i] <- median(as.matrix(M1[2:6, (1 + i):(5 + i)]))
matrixtest3[i] <- median(as.matrix(M1[3:7, (1 + i):(5 + i)]))
matrixtest4[i] <- median(as.matrix(I[4:8, (1 + i):(5 + i)]))


M1[4, 4 + seq_along(x)] <- matrixtest2[seq_along(x)]
M1[5, 5 + seq_along(x)] <- matrixtest3[seq_along(x)]
M1[6, 6 + seq_along(x)] <- matrixtest4[seq_along(x)]

This approach is obviously undesirable - any insights on how to do this, either with a for loop or apply statement would be most appreciated, thanks.

Solution

Using a double for loop.

> n <- 5L  ## define block length
> res <- vector('list', (n - 1L)^2)  ## initialize result vector
> for (i in seq_len(nrow(M1) - n + 1) - 1L) {
+   for (j in seq_len(nrow(M1) - n + 1) - 1L) {
+     tmp <- M1[1:n + i, 1:n + j]
+     res[[i + 1L]][[j + 1L]] <- replace(tmp, ceiling(length(tmp)/2), median(tmp))
+   }
+ }
> 
> res[[1]][[1]]
     [,1] [,2] [,3] [,4] [,5]
[1,]    1    4    1    4    7
[2,]    5    2    5    9    2
[3,]    1    8    4    3    5
[4,]    9    3    6    5    6
[5,]    4    1    6    8    3
> 
> identical(res[[1]][[1]], replace(M1[1:5, 1:5], 13, median(M1[1:5, 1:5])))
[1] TRUE

Data:

set.seed(42)
M1 <- matrix(sample(1:9, 400, replace=TRUE), nrow=20, ncol=20)