ivan@155: % K-SVD algorithm for Dictionary Learning
ivan@155: % Y = input data (M X L matrix)
ivan@155: % Phi = initial dictionary (M X N), e.g. random dictionary or first N data samples
ivan@155: % lambda = regularization coefficient (||Phi*X-Y||_F)^2 + lambda*||X||_1
ivan@155: % IT = number of iterations
ivan@155: function [Phiout,X,ert] = ksvd_cn(Y,Phi,lambda,IT)
ivan@155: maxIT = 1000;
ivan@155: [PhiN,PhiM] = size(Phi);
ivan@155: RR1 = PhiM;
ivan@155: %%%%%%%%%%%%%%
ivan@155: % [PhiM,L] = size(ud);
ivan@155: [PhiN,L] = size(Y);
ivan@155: X = ones(PhiM,L);
ivan@155: for it = 1:IT
ivan@155:     to = .1+svds(Phi,1);
ivan@155:     [PhiN,PhiM] = size(Phi);
ivan@155:     %%%%
ivan@155:     eps = 3*10^-4;    
ivan@155:     map = 1; % Projecting on the selected space (0=no,1=yes)
ivan@155:     [X,l1err] = mm1(Phi,Y,X,to,lambda,maxIT,eps,map); %% Sparse approximation with Iterative Soft-thresholding
ivan@155:     ert(it) = l1err;
ivan@155:     %%%          
ivan@155:     [Phi,X] = dict_update_KSVD_cn(Phi,Y,X);  
ivan@155: end
ivan@155: Phiout = Phi;