• 欢迎访问开心洋葱网站,在线教程,推荐使用最新版火狐浏览器和Chrome浏览器访问本网站,欢迎加入开心洋葱 QQ群
  • 为方便开心洋葱网用户,开心洋葱官网已经开启复制功能!
  • 欢迎访问开心洋葱网站,手机也能访问哦~欢迎加入开心洋葱多维思维学习平台 QQ群
  • 如果您觉得本站非常有看点,那么赶紧使用Ctrl+D 收藏开心洋葱吧~~~~~~~~~~~~~!
  • 由于近期流量激增,小站的ECS没能经的起亲们的访问,本站依然没有盈利,如果各位看如果觉着文字不错,还请看官给小站打个赏~~~~~~~~~~~~~!

模式识别实现之人脸识别(matlab)

人工智能 永望 2537次浏览 0个评论

1.描述 用有监督学习机制设计并实现模式识别方法,用于进行人脸面部特征识别,如性别(男性、女性)、年龄(儿童、青少年、成年、老年)、佩戴眼镜(是、否)、戴帽子(是、否)、表情(微笑、严肃)等。   2.数据 数据来源:链接:https://pan.baidu.com/s/17zU6A4-8Zs-TjspizbAOhA 提取码:dpb1   3.实现过程  
模式识别实现之人脸识别(matlab)   4.特征提取
模式识别实现之人脸识别(matlab)
模式识别实现之人脸识别(matlab) 程序流程图:
模式识别实现之人脸识别(matlab) 5. 特征提取代码  

function [dataFaceR_maleExtra,dataFaceR_femaleExtra,dataFaceS_maleExtra,dataFaceS_femaleExtra] = sexDistanceJ2(dataFaceR_male,dataFaceR_female,dataFaceS_male,dataFaceS_female,dimension)
%UNTITLED2 此处显示有关此函数的摘要
%{
    函数输入:各个类别的数据,特征提取的维数dimension
    函数的输出:经过特征提取后的各类别数据
    函数的功能:基于可分性依据J2进行特征的提取
%}
%   此处显示详细说明
[maleRow,maleCol] = size(dataFaceR_male);%获取dataFaceR_male的行数和列数
[femaleRow,femaleCol] = size(dataFaceR_female);%获取dataFaceR_female的行数和列数
%求类内的均值向量
%maleMean = mean(dataFaceR_male(:,1:maleCol+1)); %男性的均值向量
maleMean = mean(dataFaceR_male); %男性的均值向量
%femaleMean = mean(dataFaceR_female(:,1:femaleCol+1));%女性的均值向量
femaleMean = mean(dataFaceR_female);%女性的均值向量
%类别先验概率
maleP = 0.5; %男性先验概率
femaleP = 0.5;%女性先验概率
sexTotalMean = maleMean.*maleP + femaleMean.*femaleP;  %性别总体均值
%求类间离散度矩阵
   discreteSbMale = (maleMean - sexTotalMean)' * (maleMean - sexTotalMean);
 discreteSbFemale = (femaleMean - sexTotalMean)' * (femaleMean - sexTotalMean);
discreteSb = discreteSbMale.*maleP + discreteSbFemale.*femaleP;
%求类内离散度矩阵
maleDiscrete = zeros(99,99);
femaleDiscrete = zeros(99,99);
for i=1:maleRow
    maleDiscrete = maleDiscrete + (dataFaceR_male(i,:) - maleMean(1,:))' * (dataFaceR_male(i,:) - maleMean(1,:));  
end
maleDiscreteSw = (maleDiscrete.*maleP)./maleRow;     %计算男性类的类内的离散度
for i=1:femaleRow
    femaleDiscrete = femaleDiscrete + (dataFaceR_female(i,:) - femaleMean(1,:))' * (dataFaceR_female(i,:) - femaleMean(1,:));   
end
femaleDiscreteSw = (femaleDiscrete.*femaleP)./femaleRow; %计算女性类的类内的离散度
discreteSw = maleDiscreteSw + femaleDiscreteSw;%类内总的离散度
%求变换矩阵
transforMatrix = discreteSw\discreteSb;
%求变换矩阵的特征值瑜特征向量
[featureVec,eigenValMat] = eig(transforMatrix);  %featureVec:特征向量,eigenValMat:对角矩阵
eigenVal = diag(eigenValMat);    %取对角矩阵的元素,组成一列
%求特征值矩阵的行数和列数
[eigenValRow,eigenValCol] = size(eigenVal);
vec = zeros(eigenValRow,1);        %做为中间变量进行排序
Eig = zeros(1,1);                           %作为中间变量对特征值从大到小排序
%将特征值从大到小排序并跟着调制特征向量。
for i=1:eigenValRow
    k = eigenValRow - i;
    for j=1:k
        Eig = eigenVal(j,1);
        vec(:,1) = featureVec(:,j);
        if(eigenVal(j+1,1)>=Eig)
            eigenVal(j,1) = eigenVal(j+1,1);
            featureVec(:,j) = featureVec(:,j+1);
            eigenVal(j+1,1) = Eig;
            featureVec(:,j+1) = vec(:,1); 
        end    
    end
end
%获得降到dimension维数的变换矩阵
extraMatrix = featureVec(:,1:dimension);
% extraMatrix = featureVec(:,1:50);
for i=1:maleRow
    dataFaceR_maleExtra1 = dataFaceR_male(i,:)*extraMatrix;  
    dataFaceR_maleExtra(i,:) = dataFaceR_maleExtra1;     
end
for j=1:femaleRow
    dataFaceR_femaleExtra1 = dataFaceR_female(j,:)*extraMatrix;
    dataFaceR_femaleExtra(j,:) = dataFaceR_femaleExtra1;
end
%=================================================================================================================
[maleRowS,maleColS] = size(dataFaceS_male);%获取dataFaceR_male的行数和列数
[femaleRowS,femaleColS] = size(dataFaceS_female);%获取dataFaceR_female的行数和列数
%求类内的均值向量
maleMeanS = mean(dataFaceS_male); %男性的均值向量
femaleMeanS = mean(dataFaceS_female);%女性的均值向量
%类别先验概率
malePS = 0.5; %男性先验概率
femalePS = 0.5;%女性先验概率
sexTotalMeanS = maleMeanS.*malePS + femaleMeanS.*femalePS;  %性别总体均值
%求类间离散度矩阵
   discreteSbMaleS = (maleMeanS - sexTotalMeanS)' *(maleMeanS - sexTotalMeanS) ;

   discreteSbFemaleS = (femaleMeanS - sexTotalMeanS)' *(femaleMeanS - sexTotalMeanS);

discreteSbS = discreteSbMaleS.*malePS + discreteSbFemaleS.*femalePS;
%求类内离散度矩阵
maleDiscreteS = zeros(99,99);
femaleDiscreteS = zeros(99,99);
for i=1:maleRowS  
    maleDiscreteS = maleDiscreteS + (dataFaceS_male(i,:) - maleMeanS(1,:))'*(dataFaceS_male(i,:) - maleMeanS(1,:));  
end
maleDiscreteSwS = (maleDiscreteS.*malePS)./maleRowS;     %计算男性类的类内的离散度
for i=1:femaleRowS
    femaleDiscreteS = femaleDiscreteS  + (dataFaceS_female(i,:) - femaleMeanS(1,:))' * (dataFaceS_female(i,:) - femaleMeanS(1,:));  
end
femaleDiscreteSwS = (femaleDiscreteS.*femalePS)./femaleRowS; %计算女性类的类内的离散度
discreteSwS = maleDiscreteSwS + femaleDiscreteSwS;%类内总的离散度
%求变换矩阵
% ivdiscreteSwS = inv(discreteSwS);             %Sw矩阵的逆
% transforMatrixS = ivdiscreteSwS .* discreteSbS;
%求变换矩阵
transforMatrixS = discreteSwS\discreteSbS;
%求变换矩阵的特征值瑜特征向量
[featureVecS,eigenValMatS] = eig(transforMatrixS);  %featureVec:特征向量,eigenValMat:对角矩阵
eigenValS = diag(eigenValMatS);    %取对角矩阵的元素,组成一列
%求特征值矩阵的行数和列数
[eigenValRowS,eigenValColS] = size(eigenValS);
vecS = zeros(eigenValRowS,1);        %做为中间变量进行排序
EigS = 0;                           %作为中间变量对特征值从大到小排序
%将特征值从大到小排序并跟着调制特征向量。
for i=1:eigenValRowS
    for j=1:eigenValRowS - i
        EigS = eigenValS(j,1);
        vecS(:,1) = featureVecS(:,j);
        if(eigenValS(j+1,1)>=EigS)
            eigenValS(j,1) = eigenValS(j+1,1);
            featureVecS(:,j) = featureVecS(:,j+1);
            eigenValS(j+1,1) = EigS;
            featureVecS(:,j+1) = vecS(:,1); 
        end    
    end
end
%获得降到dimension维数的变换矩阵
 extraMatrixS = featureVecS(:,1:dimension);
% extraMatrixS = featureVecS(:,1:50);
for i=1:maleRowS               
    dataFaceS_maleExtra1 =dataFaceS_male(i,:) * extraMatrixS;  
    dataFaceS_maleExtra(i,:) = dataFaceS_maleExtra1;      
end
for j=1:femaleRowS
    dataFaceS_femaleExtra1 =dataFaceS_female(j,:) * extraMatrixS;  
    dataFaceS_femaleExtra(j,:) = dataFaceS_femaleExtra1;     
end
end

6.算法

function [predictSex_label, accuracySex, dec_valuesSex,lable_sexR,lable_sexS,dataFaceR_sexKnn,dataFaceS_sexKnn,label_sexRKnn,label_sexSKnn] = sex(dataFaceR_maleExtra,dataFaceR_femaleExtra,dataFaceS_maleExtra,dataFaceS_femaleExtra)
%UNTITLED 此处显示有关此函数的摘要
%   此处显示详细说明
dataFaceR_sex =[dataFaceR_maleExtra;dataFaceR_femaleExtra];
dataFaceR_sexKnn =dataFaceR_sex';

dataFaceR_sex = real(dataFaceR_sex);

[RsexRow,RsexCol] =size(dataFaceR_sex);
dataFaceS_sex = [dataFaceS_maleExtra;dataFaceS_femaleExtra];
dataFaceS_sexKnn =dataFaceS_sex';

dataFaceS_sex = real(dataFaceS_sex);

[SsexRow,SsexCol] = size(dataFaceS_sex);
[maleRow,maleCol] = size(dataFaceR_maleExtra);
[femaleRow,femaleCol] = size(dataFaceR_femaleExtra);
[maleRowS,maleColS] = size(dataFaceS_maleExtra);
[femaleRowS,femaleColS] = size(dataFaceS_femaleExtra);

lable_sexR = zeros(RsexRow,1);
lable_sexS = zeros(SsexRow,1);
for i=1:RsexRow
    if(i<=maleRow)
       lable_sexR(i,1) = 1;     %男性的标签为1,女性的标签为0
    elseif(i>maleRow)
       lable_sexR(i,1) = 2;
    end 

end
label_sexRKnn = lable_sexR';
for i=1:SsexRow
    if(i<=maleRowS)
       lable_sexS(i,1) = 1;     %男性的标签为1,女性的标签为0
    elseif(i>maleRowS)
        lable_sexS(i,1)=2;
    end 
end
label_sexSKnn = lable_sexS';

%==============================SVM==========================================================
 t1=cputime;
model = svmtrain(lable_sexR,dataFaceR_sex);
 [predictSex_label, accuracySex, dec_valuesSex] = svmpredict(lable_sexS, dataFaceS_sex, model);  % test the trainingdat
 t2=cputime;
t=t2-t1
 %===============================随机森林===================================================
 nTree = 20;  
 t1=cputime;
B = TreeBagger(nTree,dataFaceR_sex,lable_sexR);  
predict_label = predict(B,dataFaceS_sex);  
t2=cputime;
t=t2-t1
%predict_label为待转换的cell
for n=1:length(predict_label)
      x{n}=str2num(predict_label{n});
end
for m=1:length(x)
      y(m)=x{m}(1);
end
%y即为所得double类型数据
y=y';
a=0;
for i=1:SsexRow
    if(y(i,1)==lable_sexS(i,1))
        a=a+1;
    end
end
 accuracy = a/SsexRow


 %=======================================贝叶斯========================================
  t1=cputime;
 nb =fitcnb(dataFaceR_sex,lable_sexR);  
 predict_label   =  predict(nb,dataFaceS_sex);  
 t2=cputime;
t=t2-t1
a=0;
for i=1:SsexRow
    if(predict_label(i,1)==lable_sexS(i,1))
        a=a+1;
    end
end
 accuracy = a/SsexRow

%==================================集成方法============================================
% ens = fitensemble(dataFaceR_sex,lable_sexR,'AdaBoostM1' ,100,'tree','type','classification');
% predict_label   =  predict(ens, dataFaceS_sex);
% a=0;
% for i=1:SsexRow
%     if(predict_label(i,1)==lable_sexS(i,1))
%         a=a+1;
%     end
% end
%  accuracy = a/SsexRow

%================================鉴别分析分类器===========================================

% obj = ClassificationDiscriminant.fit(dataFaceR_sex,lable_sexR);
% predict_label   =       predict(obj, dataFaceS_sex);
% a=0;
% for i=1:SsexRow
%     if(predict_label(i,1)==lable_sexS(i,1))
%         a=a+1;
%     end
% end
%  accuracy = a/SsexRow
 %============================KNN===================================================
   t1=cputime;
 mdl = ClassificationKNN.fit(dataFaceR_sex,lable_sexR,'NumNeighbors',50);
  predict_label   =       predict(mdl, dataFaceS_sex);
 t2=cputime;
t=t2-t1
  a=0;
for i=1:SsexRow
    if(predict_label(i,1)==lable_sexS(i,1))
        a=a+1;
    end
end
 accuracy = a/SsexRow
end


开心洋葱 , 版权所有丨如未注明 , 均为原创丨未经授权请勿修改 , 转载请注明模式识别实现之人脸识别(matlab)
喜欢 (0)

您必须 登录 才能发表评论!

加载中……