叙述性说明

所谓的性别识别推断检测到的面部是男性还是女性。它是一个二值分类问题。

识别算法可以用于SVM，BP神经网络。LDA，PCA，PCA+LDA等等。OpenCV官网给出的文档是基于Fisherfaces检測器（LDA）方法实现的。链接：http://docs.opencv.org/modules/contrib/doc/facerec/tutorial/facerec_gender_classification.html#id5 。这篇博文（http://www.bytefish.de/blog/gender_classification/）中也是採用OpenCV官网的方法。据称有98%的正确率，我在百度图片找了一些数据測试了下，仅仅有大概50%多的识别率。原因是他的数据集是经过严格标定的。好像是眼睛的是对齐的。

实际应用中不太可能会遇到这样的情况吧。CSDN还有两篇博客也介绍到这个性格识别（http://blog.csdn.net/kklots/article/details/8247738 http://blog.csdn.net/kklots/article/details/9285505）文章写得非常好，一看就是大牛。博文中也是測试了LDA的方法，正确率也是出奇的低。採用的是PCA+LDA的方法。通过改进能达到接近90%的正确率。

博文指出PCA+LDA比单纯的LDA和PCA识别率都高，但我对博文中的PCA+LDA程序和官网的PCA程序測试了下，发现PCA的正确率会高那么一两个点。难道又是数据的问题？

数据

採集数据一方面能够採用开源的人脸库。还有一方面能够自己去百度图片下载图片。去百度或谷歌图片分别搜索“男明星头像”“女明星头像”的关键字批量下载。这里当然须要批量下载利器。

然后利用人脸检測器过滤检測出头像，然后归一化检測出来的图像，保存在本地。这样主要的数据集就有了。

当然我也会附上我採集的数据和工程文件（特此声明，全部图片均来自网络）

測试程序

创建CSV文件的python代码:

import sys

import os.path

if __name__ == "__main__":

    if len(sys.argv) != 3:

        print "usage: create_csv <base_path> <SAVE_FILE_NAME>"

        sys.exit(1)

    BASE_PATH=sys.argv[1]

    FILE_NAME = sys.argv[2]

    SEPARATOR=";"

    fh = open(FILE_NAME,'w')

    label = 0

    for dirname, dirnames, filenames in os.walk(BASE_PATH):

        for subdirname in dirnames:

            subject_path = os.path.join(dirname, subdirname)

            for filename in os.listdir(subject_path):

                abs_path = "%s/%s" % (subject_path, filename)

                ##print "%s%s%d" % (abs_path, SEPARATOR, label)

                ##print "%s  %s" % (dirname, subject_path)

                fh.write(abs_path)

                fh.write(SEPARATOR)

                if dirname.find("female") > 0 :

                    label = 1

                else:

                    label = 0

                fh.write(str(label))

                fh.write("\n")

    fh.close()

測试性别识别的程序

// gender.cpp : 定义控制台应用程序的入口点。

//

#include "stdafx.h"

#include <opencv2/opencv.hpp>

#include <iostream>

#include <fstream>

#include <sstream>

#include <math.h>

int g_howManyPhotoForTraining   =   260;

//每一个人取出8张作为训练

int g_photoNumberOfOnePerson    =   279;

//ORL数据库每一个人10张图像

using namespace cv;

using namespace std;

static Mat norm_0_255(InputArray _src) {

    Mat src = _src.getMat();

    // 创建和返回一个归一化后的图像矩阵:

    Mat dst;

    switch(src.channels()) {

case1:

        cv::normalize(_src, dst, 0,255, NORM_MINMAX, CV_8UC1);

        break;

case3:

        cv::normalize(_src, dst, 0,255, NORM_MINMAX, CV_8UC3);

        break;

    default:

        src.copyTo(dst);

        break;

    }

    return dst;

}

//使用CSV文件去读图像和标签，主要使用stringstream和getline方法

static void read_csv(const string& filename, vector<Mat>& images, vector<int>& labels, char separator =';') {

    std::ifstream file(filename.c_str(), ifstream::in);

    if (!file) {

        string error_message ="No valid input file was given, please check the given filename.";

        CV_Error(CV_StsBadArg, error_message);

    }

    string line, path, classlabel;

    while (getline(file, line)) {

        stringstream liness(line);

        getline(liness, path, separator);

        getline(liness, classlabel);

        if(!path.empty()&&!classlabel.empty()) {

            images.push_back(imread(path, 0));

            labels.push_back(atoi(classlabel.c_str()));

        }

    }

}

void train_and_test_lda()

{

    string fn_csv = string("at.txt");

    //string fn_csv = string("feret.txt");

    vector<Mat> allImages,train_images,test_images;

    vector<int> allLabels,train_labels,test_labels;

    try {

        read_csv(fn_csv, allImages, allLabels);

    } catch (cv::Exception& e) {

        cerr <<"Error opening file "<< fn_csv <<". Reason: "<< e.msg << endl;

        // 文件有问题，我们啥也做不了了。退出了

        exit(1);

    }

    if(allImages.size()<=1) {

        string error_message ="This demo needs at least 2 images to work. Please add more images to your data set!";

        CV_Error(CV_StsError, error_message);

    }

    for(int i=0 ; i<allImages.size() ; i++)

        equalizeHist(allImages[i],allImages[i]);

    int photoNumber = allImages.size();

    for(int i=0 ; i<photoNumber ; i++)

    {

        if((i%g_photoNumberOfOnePerson)<g_howManyPhotoForTraining)

        {

            train_images.push_back(allImages[i]);

            train_labels.push_back(allLabels[i]);

        }

        else

        {

            test_images.push_back(allImages[i]);

            test_labels.push_back(allLabels[i]);

        }

    }

    /*Ptr<FaceRecognizer> model = createEigenFaceRecognizer();//定义pca模型

    model->train(train_images, train_labels);//训练pca模型。这里的model包括了全部特征值和特征向量。没有损失

    model->save("eigenface.yml");//保存训练结果。供检測时使用  */

    Ptr<FaceRecognizer> fishermodel = createFisherFaceRecognizer();

    fishermodel->train(train_images,train_labels);//用保存的降维后的图片来训练fishermodel，后面的内容与原始代码就没什么变化了

    fishermodel->save("fisherlda.yml");

    int iCorrectPrediction = 0;

    int predictedLabel;

    int testPhotoNumber = test_images.size();

    for(int i=0;i<testPhotoNumber;i++)

    {

        predictedLabel = fishermodel->predict(test_images[i]);

        if(predictedLabel == test_labels[i])

            iCorrectPrediction++;

    }

    string result_message = format("Test Number = %d / Actual Number = %d.", testPhotoNumber, iCorrectPrediction);

    cout << result_message << endl;

    cout<<"accuracy = "<<float(iCorrectPrediction)/testPhotoNumber<<endl;

}

void train_and_test_pca()

{

    string fn_csv = string("at.txt");

    //string fn_csv = string("feret.txt");

    vector<Mat> allImages,train_images,test_images;

    vector<int> allLabels,train_labels,test_labels;

    try {

        read_csv(fn_csv, allImages, allLabels);

    } catch (cv::Exception& e) {

        cerr <<"Error opening file "<< fn_csv <<". Reason: "<< e.msg << endl;

        // 文件有问题。我们啥也做不了了，退出了

        exit(1);

    }

    if(allImages.size()<=1) {

        string error_message ="This demo needs at least 2 images to work. Please add more images to your data set!";

        CV_Error(CV_StsError, error_message);

    }

    for(int i=0 ; i<allImages.size() ; i++)

        equalizeHist(allImages[i],allImages[i]);

    int photoNumber = allImages.size();

    for(int i=0 ; i<photoNumber ; i++)

    {

        if((i%g_photoNumberOfOnePerson)<g_howManyPhotoForTraining)

        {

            train_images.push_back(allImages[i]);

            train_labels.push_back(allLabels[i]);

        }

        else

        {

            test_images.push_back(allImages[i]);

            test_labels.push_back(allLabels[i]);

        }

    }

    Ptr<FaceRecognizer> model = createEigenFaceRecognizer();//定义pca模型

    model->train(train_images, train_labels);//训练pca模型，这里的model包括了全部特征值和特征向量。没有损失

    model->save("eigenfacepca.yml");//保存训练结果，供检測时使用

    int iCorrectPrediction = 0;

    int predictedLabel;

    int testPhotoNumber = test_images.size();

    for(int i=0;i<testPhotoNumber;i++)

    {

        predictedLabel = model->predict(test_images[i]);

        if(predictedLabel == test_labels[i])

            iCorrectPrediction++;

    }

    string result_message = format("Test Number = %d / Actual Number = %d.", testPhotoNumber, iCorrectPrediction);

    cout << result_message << endl;

    cout<<"accuracy = "<<float(iCorrectPrediction)/testPhotoNumber<<endl;

}

void train_and_test()

{

    string fn_csv = string("at.txt");

    //string fn_csv = string("feret.txt");

    vector<Mat> allImages,train_images,test_images;

    vector<int> allLabels,train_labels,test_labels;

    try {

        read_csv(fn_csv, allImages, allLabels);

    } catch (cv::Exception& e) {

        cerr <<"Error opening file "<< fn_csv <<". Reason: "<< e.msg << endl;

        // 文件有问题，我们啥也做不了了。退出了

        exit(1);

    }

    if(allImages.size()<=1) {

        string error_message ="This demo needs at least 2 images to work. Please add more images to your data set!";

        CV_Error(CV_StsError, error_message);

    }

    for(int i=0 ; i<allImages.size() ; i++)

        equalizeHist(allImages[i],allImages[i]);

    int photoNumber = allImages.size();

    for(int i=0 ; i<photoNumber ; i++)

    {

        if((i%g_photoNumberOfOnePerson)<g_howManyPhotoForTraining)

        {

            train_images.push_back(allImages[i]);

            train_labels.push_back(allLabels[i]);

        }

        else

        {

            test_images.push_back(allImages[i]);

            test_labels.push_back(allLabels[i]);

        }

    }

    Ptr<FaceRecognizer> model = createEigenFaceRecognizer();//定义pca模型

    model->train(train_images, train_labels);//训练pca模型。这里的model包括了全部特征值和特征向量，没有损失

    model->save("eigenface.yml");//保存训练结果。供检測时使用

    Mat eigenvalues = model->getMat("eigenvalues");//提取model中的特征值。该特征值默认由大到小排列

    Mat W = model->getMat("eigenvectors");//提取model中的特征向量，特征向量的排列方式与特征值排列顺序一一相应

    int xth = 121;//打算保留前121个特征向量，代码中没有体现原因。但选择121是经过斟酌的，首先，在我的实验中。"前121个特征值之和/全部特征值总和>0.97"。其次，121=11^2，能够将结果表示成一个11*11的2维图像方阵，交给fisherface去计算。

vector<Mat> reduceDemensionimages;//降维后的图像矩阵

    vector<Mat> testreduceDemensionimages;

    Mat evs = Mat(W, Range::all(), Range(0, xth));//选择前xth个特征向量，其余舍弃

    Mat mean = model->getMat("mean");

    for(int i=0;i<train_images.size();i++)

    {

        Mat projection = subspaceProject(evs, mean, train_images[i].reshape(1,1));//做子空间投影

        reduceDemensionimages.push_back(projection.reshape(1,sqrt(xth*1.0)));//将获得的子空间系数表示映射成2维图像，并保存起来

    }  

    for(int i=0;i<test_images.size();i++)

    {

        Mat projection = subspaceProject(evs, mean, test_images[i].reshape(1,1));//做子空间投影

        testreduceDemensionimages.push_back(projection.reshape(1,sqrt(xth*1.0)));//将获得的子空间系数表示映射成2维图像。并保存起来

    }

    Ptr<FaceRecognizer> fishermodel = createFisherFaceRecognizer();

    fishermodel->train(reduceDemensionimages,train_labels);//用保存的降维后的图片来训练fishermodel。后面的内容与原始代码就没什么变化了

    fishermodel->save("fisher.yml");

    int iCorrectPrediction = 0;

    int predictedLabel;

    int testPhotoNumber = test_images.size();

    for(int i=0;i<testPhotoNumber;i++)

    {

        predictedLabel = fishermodel->predict(testreduceDemensionimages[i]);

        if(predictedLabel == test_labels[i])

            iCorrectPrediction++;

    }

    string result_message = format("Test Number = %d / Actual Number = %d.", testPhotoNumber, iCorrectPrediction);

    cout << result_message << endl;

    cout<<"accuracy = "<<float(iCorrectPrediction)/testPhotoNumber<<endl;

}

void test_pca()

{

    string fn_csv = string("test.txt");

    vector<Mat> allImages;

    vector<int> allLabels;

    try {

        read_csv(fn_csv, allImages, allLabels);

    } catch (cv::Exception& e) {

        cerr <<"Error opening file "<< fn_csv <<". Reason: "<< e.msg << endl;

        // 文件有问题，我们啥也做不了了，退出了

        exit(1);

    }

    if(allImages.size()<=1) {

        string error_message ="This demo needs at least 2 images to work. Please add more images to your data set!";

        CV_Error(CV_StsError, error_message);

    }

    Ptr<FaceRecognizer> model = createEigenFaceRecognizer();//定义pca模型

    model->load("eigenfacepca.yml");//保存训练结果。供检測时使用  

    int iCorrectPrediction = 0;

    int predictedLabel;

    int testPhotoNumber = allImages.size();

    for(int i=0;i<testPhotoNumber;i++)

    {

        predictedLabel = model->predict(allImages[i]);

        if(predictedLabel == allLabels[i])

            iCorrectPrediction++;

    }

    string result_message = format("Test Number = %d / Actual Number = %d.", testPhotoNumber, iCorrectPrediction);

    cout << result_message << endl;

    cout<<"accuracy = "<<float(iCorrectPrediction)/testPhotoNumber<<endl;

}

void test()

{

    string fn_csv = string("test.txt");

    vector<Mat> allImages;

    vector<int> allLabels;

    try {

        read_csv(fn_csv, allImages, allLabels);

    } catch (cv::Exception& e) {

        cerr <<"Error opening file "<< fn_csv <<". Reason: "<< e.msg << endl;

        // 文件有问题。我们啥也做不了了，退出了

        exit(1);

    }

    if(allImages.size()<=1) {

        string error_message ="This demo needs at least 2 images to work. Please add more images to your data set!";

        CV_Error(CV_StsError, error_message);

    }

    Ptr<FaceRecognizer> model = createEigenFaceRecognizer();//定义pca模型

    model->load("eigenface.yml");//保存训练结果。供检測时使用

    Mat eigenvalues = model->getMat("eigenvalues");//提取model中的特征值，该特征值默认由大到小排列

    Mat W = model->getMat("eigenvectors");//提取model中的特征向量，特征向量的排列方式与特征值排列顺序一一相应

    int xth = 121;//打算保留前121个特征向量，代码中没有体现原因，但选择121是经过斟酌的，首先，在我的实验中。"前121个特征值之和/全部特征值总和>0.97"；其次，121=11^2，能够将结果表示成一个11*11的2维图像方阵，交给fisherface去计算。

    vector<Mat> reduceDemensionimages;//降维后的图像矩阵

    Mat evs = Mat(W, Range::all(), Range(0, xth));//选择前xth个特征向量。其余舍弃

    Mat mean = model->getMat("mean");

    for(int i=0;i<allImages.size();i++)

    {

        Mat projection = subspaceProject(evs, mean, allImages[i].reshape(1,1));//做子空间投影

        reduceDemensionimages.push_back(projection.reshape(1,sqrt(xth*1.0)));//将获得的子空间系数表示映射成2维图像，并保存起来

    }  

    Ptr<FaceRecognizer> fishermodel = createFisherFaceRecognizer();  

    fishermodel->load("fisher.yml");

    int iCorrectPrediction = 0;

    int predictedLabel;

    int testPhotoNumber = allImages.size();

    for(int i=0;i<testPhotoNumber;i++)

    {

        predictedLabel = fishermodel->predict(reduceDemensionimages[i]);

        if(predictedLabel == allLabels[i])

            iCorrectPrediction++;

    }

    string result_message = format("Test Number = %d / Actual Number = %d.", testPhotoNumber, iCorrectPrediction);

    cout << result_message << endl;

    cout<<"accuracy = "<<float(iCorrectPrediction)/testPhotoNumber<<endl;

}

int main() {

    cout<<"lda = "<<endl;

    train_and_test_lda();

    cout<<"pca = "<<endl;

    train_and_test_pca();

    cout<<"pca+lda = "<<endl;

    train_and_test();

    /*test();

    test_pca();*/

    return 0 ;

}

整个工程文件和数据下载链接 http://download.csdn.net/detail/zwhlxl/8510649

版权声明：本文博主原创文章，博客，未经同意不得转载。