程序分析和总结.docx

读图像并构成训练样本和测试样本train_xulie=[];test_xulie=[];

fori=1:10

fname=strcat(E:\face\Yaletrain\,int2str(5),Train,\,int2str(i),.mat);

load(fname);train_xulie=[train_xulie,trainIdx];test_xulie=[test_xulie,testIdx];

endcleari;

loadE:\face\Yale_32x32.mat;facedata=fea;trainIndex=train_xulie;testIndex=train_xulie;

Class_train_num=5;%每个类别中的训练样本数目

Class_num=15;%类别数目

Class_test_num=11-(Class_train_num);Train_num=Class_num*Class_train_num;Test_num=Class_num*Class_test_num;fea=facedata;

index=1;

%%%%%%%%%%%%readingtrainimage %%%%%%%%%%%%%%%%%%%%%%fori=1:Class_num

forj=1:Class_train_num

index=(i-1)*Class_train_num+j;a=fea(trainIndex(index),:);train_matrix(index,:)=a;

cleara;clearb;clearc;

end

end

cleari;clearj;

TA=train_matrix;%训练样本向量矩阵,行数为样本个数,每个行向量为一个样本

%%%%%%%%%%%%%%%readingtestimages %%%%%%%%%%%%%%%%%%%clearindex;

cleart;

fori=1:Class_num

forj=1:Class_test_num

index=(i-1)*Class_test_num+j;t=fea(testIndex(index),:);

Test_image(index,:)=t;

end

end

cleart;clearb;clearc;

TE=Test_image;%测试样本向量矩阵,行数为样本个数,每个行向量为一个样本

%%%%%%%%%%%%%%%%%%%%%根据训练样本构造支持向量,采用决策数算法%%%%%%%%%%%%%%%%%%

C =4; %4;

ker=rbf;

p = 0.5; %0.5;

global p1 ; % kernel func s argument. RBF: k =exp(-(u-v)*(u-v)/(2*p1^2));POLY:k=(u*v+1)^p1;

globalp2 ; %kernelfuncsargument.p1=p;

d=4;

fori=1:(Class_num-1)

Y1=-1*ones(((Class_num+1)-i)*Class_train_num,1);forj=1:Class_train_num

Y1(j)=1;

end

X1 =

TA((i*Class_train_num-(Class_train_num-1)):(Class_num*Class_train

_num),:);

[nsv,alpha,b0]=svc(X1,Y1,ker,C,d);

%保存信息

filename=sprintf(e:\\zhichi\\svm-%d.bin,i);hg=fopen(filename,w+);o=fwrite(hg,Y1,double);u=fwrite(hg,alpha,double);v=fwrite(hg,b0,double);zz=fwrite(hg,nsv,double);%生成标准支持向量.fclose(hg);

clearalpha;clearnsv;clearb0;

end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%利用支持向量机进行识别%%%%%%%%%%%%%%p1=4;

nsv=ones(Class_num-1,1);Predict=zeros(Class_num,Class_test_num);st=cputime;

PredictY=0;

fori=1:Class_num

forj=1:Class_test_num

tstx=T