parzen.m

最新的模式识别分类工具箱,希望对朋友们有用!

function D = parzen(train_features, train_targets,params, region)

% Classify using the Parzen windows algorithm
% Inputs:
% 	features  - Train features
%	targets	  - Train targets
%	param     - [hParzen, foo]  - Normalizing factor for h, unused
%	region	  - Decision region vector: [-x x -y y number_of_points]
%
% Outputs
%	D			- Decision sufrace

comma_loc = findstr(params,',');
hParzen   = str2num(params(2:comma_loc(1)-1));

N		= region(5);								%Number of points on the grid
x		= ones(N,1) * linspace (region(1),region(2),N);
y		= linspace (region(3),region(4),N)' * ones(1,N);
V0		= zeros(N);
V1		= zeros(N);

h_factor   = hParzen;
train_one  = find(train_targets == 1);
train_zero = find(train_targets == 0);


%Estimate probabilities for for class 0

%Estimate mean and covariance for class 0
m0 = mean(train_features(:,train_zero)');
s0 = cov(train_features(:,train_zero)');
P0 = length(train_zero)/length(train_features);

sigma0 = sqrt(sum(diag(s0)));
h0 	 = sigma0/h_factor;

n		= length(train_zero);

for i = 1:n,
   if (i/50 == floor(i/50)),
      disp(['Finished ' num2str(i) ' iterations out of ' num2str(n) ' iterations.'])
   end
   temp = (x - train_features(1,train_zero(i))).^2 + (y - train_features(2,train_zero(i))).^2;
   V0   = V0 + exp(-temp./(2*h0^2));
end

V0 = 1/(sqrt(2*pi)*h0^2)*V0/n;


%Estimate mean and covariance for class 1
m1 = mean(train_features(:,train_one)');
s1 = cov(train_features(:,train_one)');
P1 = length(train_one)/length(train_features);

sigma1 = sqrt(sum(diag(s1)));
h1 	 = sigma1/h_factor;

n		= length(train_one);

for i = 1:n,
   if (i/50 == floor(i/50)),
      disp(['Finished ' num2str(i) ' iterations out of ' num2str(n) ' iterations.'])
   end
   temp = (x - train_features(1,train_one(i))).^2 + (y - train_features(2,train_one(i))).^2;
   V1   = V1 + exp(-temp./(2*h1^2));
end

V1 = 1/(sqrt(2*pi)*h1^2)*V1/n;

D = (V0*P0 < V1*(1-P0));