Picking values from n-d array along 3rd dimension

Hi,

Suppose I have A, an m x n x q array with large integer values.

I have B, an m x n array of integer values in the range of 1:q.

I'd like to pick values from A along the 3rd dimension, so that I obtain C, an m x n array in which for every subscript pair (i,j), m>i>1, n>j>1 the following is true:

x = B(i,j)
C(i,j) = A(i,j,x)

Is there an efficient, loop-free way of doing this?

Many thanks,

Answer 1

[Andrei Bobrov]

[ii,jj] = ndgrid(1:m,1:n);
ijk = sub2ind(size(A),ii,jj,B);
C = A(ijk);

[Pal]

Thank you!

Answer 2

[Roger Stafford]

 C = A((1:m*n)'+m*n*(B(:)-1));
 C = reshape(C,m,n);

[Andrei Bobrov]

 A(reshape(1:m*n,m,[]) + m*n*(B-1))

[Pal]

This is 2X faster than the accepted answer. Thanks!

[Roger Stafford]

A(reshape(1-m*n:0,m,[])+m*n*B)

Answer 3

[David Sanchez]

You can assign the value of C without referencing (i,j):

x = B(i,j);
C = A(:,:,x);

And you will have for all i, j that C(i,j) = A(i,j,x)

[Pal]

Could you elaborate, please? If I write

C = A(:,:,B),

the answer is an m x n x numel(B) array, which is incorrect. What do you mean by 'without referencing (i,j)'? B is not constant.