k_means.asv

a matlab program that deals with k-means algorithm to cluster the data points and build codebook.

function centroid = k_means(X)

% figure(1);
plot(X(1,:),X(2,:),'.');title('Initial Data');

numClass = 5; % number of classes should be provided to the algorithm
centroid(:,1:numClass)=X(:,1:numClass); % centroid initialization

n1 = 0; % number of data in class1
n2 = 0;
n3 = 0;
n4 = 0;
n5 = 0;

class1 = zeros(2,1);% class initialization
class2 = zeros(2,1);
class3 = zeros(2,1);
class4 = zeros(2,1);
class5 = zeros(2,1);
centroid1 = [0;0]; % centroid1 is old_centroid

while abs(centroid(:,1)-centroid1)>0.001,% criteria for the loop to stop
    % it checks the difference between old centroid and update centroid
    X1=[centroid, X]; %embedding centroids to training data
    distMatrix = squareform(pdist(X1')); % finding the euclidean distance matrix
    distMatrix = distMatrix(1:numClass, numClass+1:end); % 
    [junk index] = min(distMatrix);

    for m = 1:size(distMatrix, 2),
        if index(m) == 1,
            n1=n1+1;
            class1(:,n1) = X(:,m);
        elseif index(m)==2,
            n2 = n2+1;
            class2(:,n2) = X(:,m);
        elseif index(m)==3,
            n3 = n3+1;
            class3(:,n3) = X(:,m);
        elseif index(m)==4,
            n4 = n4+1;
            class4(:,n4) = X(:,m);
        elseif index(m)==5,
            n5 = n5+1;
            class5(:,n5) = X(:,m);
        end
    end

    centroid1 = centroid(:,1);

    centroid(:,1) = mean(class1,2);
    centroid(:,2) = mean(class2,2);
    centroid(:,3) = mean(class3,2);
    centroid(:,4) = mean(class4,2);
    centroid(:,5) = mean(class5,2);
end

% figure;
plot(class1(1,:),class1(2,:),'mo', class2(1,:),class2(2,:),'co',class3(1,:),...
    class3(2,:),'ro',class4(1,:),class4(2,:),'go',class5(1,:),class5(2,:),'bo');hold on;
legend('class1','class2','class3','class4','class5');
plot(centroid(1,:), centroid(2,:),'ks');hold off;
title('Data after classification');