Assuming that each row/col position of B has exactly one 1 throughout its depth, then this can be done like follows:
%An A
A = repmat(magic(10),[1 1 5]);
%Simulate a B where each row/col pair has exactly one true value through
%depth
B = false(size(A));
[~,idx] = max(rand(size(A)),[],3);
[rr, cc] = ndgrid(1:size(A,1),1:size(A,2));
B(sub2ind(size(B),rr(:), cc(:), idx(:))) = true;
%Check it
assert(all(all(sum(B,3)==1)))
%Now, how do we undo the above?
[~,idx] = max(B,[],3); %which page?
C =reshape(A(sub2ind(size(B),rr(:), cc(:), idx(:))),size(idx)); %rr/cc from above, reshape to original shape
%Check it
assert(isequal(C,magic(10)));
And of course doc sub2ind will be your friend.
