fastica算法代码.txt

关于人脸识别中的ICA算法

FastICA算法代码
%对数据进行标准化
close all,clear
load np_all
normsr=zscore(npp);
features=normsr';
[r,c]= size(features);
%对数据进行白化，使之为0均值，单位协方差
features = features - mean(features')'*ones(1,c);
[E, D]  = eig(cov(features',1));
V      = E*inv(sqrtm(D))*E';
z = V*features;
%使用FastICA算法对数据进行特征提取.非线性函数选择tanh(a1y)
epsilon=0.0001;
m=r;
p=1;
W=zeros(r,1);
for p=1:m
W(:,p)=rand(r,1);
W(:,p)=W(:,p)/norm(W(:,p));
exit=0;
count=0;
iter=1;
while exit==0;
count=count+1;
temp=W(:,p);
for i=1:m
W(i,p)=mean(z(i,.*(tanh((temp)'*z)))-(mean(1-(tanh((temp))'*z).^2)).*temp(i,1);
end
ssum=zeros(r,1);
for counter=1:p-1
ssum=ssum+(W(:,p)'*W(:,counter))*W(:,counter);
end
W(:,p)=W(:,p)-ssum;
W(:,p)=W(:,p)/norm(W(:,p));
if(abs((dot(W(:,p),temp)))<1+epsilon)&(abs((dot(W(:,p),temp)))>1-epsilon)
exit=1;
end
iter=iter+1;
end
end
W=W';
%找到最独立的方向
power=sum(abs(W)');
[m, in]  = sort(power);
W= W(in(r-2+1:r),;
%计算新的特征数据.
features = W*z;
features=features';
X=features(find(kk==1|kk==2),;
group=kk(find(kk==1|kk==2));
%画散点图.
gscatter(X(:,1),X(:,2),group,'mr','o+') 
legend('health','npc-initial');
xlabel('1st component');
ylabel('2nd component');