tprod.m

function C = tprod(A,B,transform)

% C = tprod(A,B,transform) computes the tensor-tensor product of two 3 way tensors A and B under linear transform
%
% Input:
%       A       -   n1*n2*n3 tensor
%       B       -   n2*m2*n3 tensor
%   transform   -   a structure which defines the linear transform
%       transform.L: the linear transform of two types:
%                  - type I: function handle, i.e., @fft, @dct
%                  - type II: invertible matrix of size n3*n3
%
%       transform.inverseL: the inverse linear transform of transform.L
%                         - type I: function handle, i.e., @ifft, @idct
%                         - type II: inverse matrix of transform.L
%
%       transform.l: a constant which indicates whether the following property holds for the linear transform or not:
%                    L'*L=L*L'=l*I, for some l>0.                           
%                  - transform.l > 0: indicates that the above property holds. Then we set transform.l = l.
%                  - transform.l < 0: indicates that the above property does not hold. Then we can set transform.l = c, for any constant c < 0.
%       If not specified, fft is the default transform, i.e.,
%       transform.L = @fft, transform.l = n3, transform.inverseL = @ifft. 
%
%
% Output: C     -   n1*m2*n3 tensor, C = A * B 
%
%
%
% See also lineartransform, inverselineartransform
%
%
% References:
% Canyi Lu, Tensor-Tensor Product Toolbox. Carnegie Mellon University. 
% June, 2018. https://github.com/canyilu/tproduct.
%
% Canyi Lu, Xi Peng, Yunchao Wei, Low-Rank Tensor Completion With a New Tensor 
% Nuclear Norm Induced by Invertible Linear Transforms. IEEE International 
% Conference on Computer Vision and Pattern Recognition (CVPR), 2019
%
% Canyi Lu, Pan Zhou. Exact Recovery of Tensor Robust Principal Component 
% Analysis under Linear Transforms. arXiv preprint arXiv:1907.08288. 2019
%
%
% version 1.0 - 01/02/2019
% version 1.1 - 29/04/2021
%
% Written by Canyi Lu (canyilu@gmail.com)
%

[n1,n2,n3] = size(A);
[m1,m2,m3] = size(B);

if n2 ~= m1 || n3 ~= m3
    error('Inner tensor dimensions must agree.');
end
if nargin < 3
    % fft is the default transform
    transform.L = @fft; transform.l = n3; transform.inverseL = @ifft;
end

C = zeros(n1,m2,n3);
if isequal(transform.L,@fft)
    % efficient computing for fft transform
    A = fft(A,[],3);
    B = fft(B,[],3);
    halfn3 = ceil((n3+1)/2);
    for i = 1 : halfn3
        C(:,:,i) = A(:,:,i)*B(:,:,i);
    end
    for i = halfn3+1 : n3
        C(:,:,i) = conj(C(:,:,n3+2-i));
    end
    C = ifft(C,[],3);
else
    % other transform
    A = lineartransform(A,transform);
    B = lineartransform(B,transform);
    for i = 1 : n3
        C(:,:,i) = A(:,:,i)*B(:,:,i);
    end
    C = inverselineartransform(C,transform);
end