Slightly simpler version of @Nora Khaled's idea. It will also generalize to an (N-1) X N array in a way that I believe makes sense, in case that is important.
M(:,i-1:i) = H(j:j+r-1,i-1:i);
You don't really need to define r or j as a separate variable, but I think it is a bit clearer.
Also, if you did need this for large arrays, it is more memory-efficient to preallocate M as I did, rather than building it by continually appending.