How to use a 2D index to create a 3D array from a 4D array?

Hello,

I have a main 4D array A with size

200x300x3x5

and I want to create a 3D array C of size

200x300x3

using a 2D array index that picks one specific set of the 4D array A to keep to make C.

I have a 2D index array B with size

200x300

, and all elements of B range from 1 to 5 (i.e. size of the 4th dimension of A). Each element in B indicates to which set of the 4th dimension element I want to access from A to use to create C.

I want to avoid using loops to reduce run time but at the same time I'm also lost as to how to do this.

I tried

C(:,:,:) = (A(B))

and

C(:,:,:) = (A(:,:,:,B))    

but neither seem to do what I want it to do.

[Guillaume]

Why is the index 2d and not 3d? Is it the same 4th dim index you want to pick for all 3 pages? I.e. should the index be repmat'ed across the 3rd dimension?

[Janette Lin]

The each of the 4th dimension is the same for all 3 pages

Answer 1

[Bruno Luong]

It's amazing you all ask the same question, it must have some competition somewhere in the world (china?) about the topic of processing a stack of RGB images.

[m,n,p,~] = size(A);
x = m*n*p;
C = reshape(A((1:x)'+x*(repmat(B(:)-1,p,1))),[m,n,p]);

[Bob Thompson]

Ha, I totally recognized the question too while I was reading it.

[Bruno Luong]

I detect at least three guys around the table. ;-)

[Janette Lin]

What does this bit in your third line of your code do?

(1:x)'+x*(repmat(B(:)-1,p,1))

[Bruno Luong]

it computes the linear-index of each voxel on the "page" B.

You might interest in Andrei's use of NDGRID and SUB2IND. This is clearer coding though it requires 4/5 big temporary arrays.

My code is then more obscure but faster.

[Andrei Bobrov]

+1