SciPy convolve2d vs Matlab conv2 'same' mode

gistfile1.txt

In [198]: f = [[2,3,4],[1,6,7]]

In [199]: g = [[9,1,0],[2,5,8],[1,3,3]]

In [200]: convolve2d(f,g)
Out[200]: 
array([[ 18,  29,  39,   4,   0],
       [ 13,  71, 108,  51,  32],
       [  4,  26,  71, 104,  68],
       [  1,   9,  28,  39,  21]])

In [201]: convolve2d(f,g,'same')
Out[201]: 
array([[ 71, 108,  51],
       [ 26,  71, 104]])

In [202]: convolve2d(g,f,'same')
Out[202]: 
array([[ 29,  39,   4],
       [ 71, 108,  51],
       [ 26,  71, 104]])

Compare with the example in http://books.google.com/books?id=fTvQkLBLyK8C&lpg=PA43&ots=0HC1LsH0cE&dq=conv2%20same%20example&pg=PA43#v=onepage&q=conv2%20same%20example&f=false

 71 108 51
 26 71 104

and

 71 108  51
 26  71 104
  9  28 39

http://www.mathworks.com/help/matlab/ref/conv2.html "If there are an odd number of rows or columns, the "center" leaves one more at the beginning than the end."