k-means算法代码和VQ图片集

function [c,mu,ex] = km(x,K) %-------------------------------------------------------------------------- % kmeans : K-means clustering % [c,mu] = kmeans(x,K) % x - d*n samples % K - number of clusters wanted % mu - optional d*K initial guess for cluster centroids % returns: % c - 1*n calculated membership vector where c(j) \in 1..K % mu - d*K cluster centroids % Check arguments if ndims(x)~=2, error('x must be a matrix'); end [d,n] = size(x); if ~isequal(size(K),[1 1]), error('K must be a scalar');end K=round(K); if (K<1 || K>n), error(sprintf('K must be at least one and at most n=%d',n)); end % compute time tic % Initialize variables oldmu = Inf*ones(d,K); c = zeros(1,n); D = zeros(K,n); % choose k from n as the cluster centers if nargin<3 p=randperm(n); mu = x(:,p(1:K)); else disp(['error!!!']); return; end % while(1) for iter = 1:100*log2(K) % for iter = 1:48*log(K)/log(2) % for iter = 1:5 % distribute the samples x to the chosen cluster domains for j=1:K, % for every cluster center = mu(:,j); % get cluster center if ~isequal(center,oldmu(:,j)) D(j,:) = sum( ( x - repmat( center, 1, n ) ).^2 ); end end % find minimum if K ~= 1 [Dmin,index] = min(D); else Dmin = D; index = ones(1,n); end % moved = sum(index~=c); c = index; % update centroid % distkm = zeros(1,K); for i=1:K ci=find(c==i); if sum(ci)~= 0 mu(:,i)=mean(x(:,ci),2); continue;end end % Check convergence % if (moved==0), break, end end % calculate distkm = sum( Dmin ); % record the results ex.distk = distkm; ex.time = toc; % compute time time_km = toc