knnclass.m

很好用的统计模式识别工具箱

function [class,index,dist] = knnclass(tst,X,I,K)
% [class,index,dist] = knnclass(tst,X,I,K)
%
% KNNCLASS is an implementation of K-Nearest Neighbours 
%  classifier. The Euclidean metric is used.
%
% Input:
%  tst [DxNtst] Ntst test points of dimension D.
%  X [D,Ntrn] Ntrn training points of dimension D.
%  I [1,Ntrn] Ntrn labels of the training points.
%  K [1x1] number of nearest neighbours.
%
% Output:
%  class [1xNtst] class labels for each test point.
%  index [1xNtst] index of the nearest point from the traing set.
%  dist [1xNtst] distance from the nearest point from the traing set.
%

% Statistical Pattern Recognition Toolbox, Vojtech Franc, Vaclav Hlavac
% (c) Czech Technical University Prague, http://cmp.felk.cvut.cz
% Written Vojtech Franc (diploma thesis)
% Modifications
% 26-feb-2001 V.Franc


ntst=size(tst,2);
ntrn=size(X,2);

nclass=max(I)-min(I)+1;

class=zeros(1,ntst);
index=zeros(1,ntst);
dist=zeros(1,ntst);

for i=1:ntst,

  d = diag((X-repmat(tst(:,i),1,ntrn))'*(X-repmat(tst(:,i),1,ntrn)));
  [d,inx]=sort(d);    % sort distances in ascending order
  
  knn=I(inx);
  
  index(i)=inx(1);   % get distance of the nearest point and its index
  dist(i)=d(1);
  
  h=hist(knn(1:K),[1:nclass]);
  
  [cnt,class(i)]=max(h);

end

dist=sqrt(dist);
  
return