Iterative Principal Component Analysis

iPCA.m

function [Un, mn, An, S, nr] = iPCA(x, U, m, A, max_dim)
% Execute incremental PCA analysis
% Input
% x - new input vector (Dx1)
% Uo - Last basis (Dxd)
% mo - last mean  (Dx1)
% Ao - last coefficients (dxn)
% max_dim - maximum number of dimensions to consider, if this not
% prescribed or null then the function augments the basis accorindgly

% Choose the preffered dimensionality 
reduce = 0;
if nargin == 5 && numel(max_dim) > 0
    reduce = 1;
end

% Initialize on first, on first iteration define U,m, and A as null
if (numel(U) + numel(A) )== 0
    Un = zeros(numel(x),1);
    An = 0;
    mn = x(:);
    S = 0;
    nr = 0;
    disp('Initializing')
    return
end

% This is from the second iteration on
x = x(:); % assure that x is a column vector
m = m(:); % assure that the mean is a column vector
% compute the new projection

if size(U,1) < size(U,2) 
    error('The dimensions of the basis are inconsistent with the algorithm');
end

a = U' * (x - m); % compute the new projection   dx1
y = U * a + m;    % Reconstruct input  Dx1
r = x - y ;          % Residual of reconstruction
nr = norm(r); 

Up = [U, r./nr];     % D x d+1
Ap =  [ A, a; zeros(1,size(A,2)), nr];    %  d+1  x n+1

% batch PCA
mpp = mean(Ap,2);  % mean of new projections  dx1
Atemp = Ap - repmat( mpp, 1, size(Ap,2));

C = Atemp * Atemp';  % Mean centered covariance matrix

% batch PCA
[Upp, S] = eig(C);

% disp([sum(sum(Up.^2,2)) sum(sum(Up.^2,1))])
An = Upp' * ( Ap - mpp * ones(1, size(Atemp,2)));  % Update coefficients

Un = Up * Upp;  % Update basis
mn = m + Up*mpp; % Update mean

if reduce && size(S,1) > max_dim
    Un(:,1) = []; S(1,:) = []; S(:,1) = []; An(1,:) = [];
end

No. I don't know if it works. I was hoping you could help me.