利用HOG+SVM训练自己的人脸检测XML文件

1、生成xml训练文件

#include<iostream>
#include<opencv2/opencv.hpp>
#include<vector>
using namespace std;
using namespace cv;
using namespace cv::ml;
#define HOG_VECTOR 900
#define TRAIN_IMG_NUM 800
int main()
{
	//创建HOG描述子
	HOGDescriptor hog(Size(48, 48), Size(16, 16), Size(8, 8), Size(8, 8), 9,-1);
	//提取样本特征
	vector<float> descrip;//每个检测窗口的特征向量
	vector<float> totaldescrip(HOG_VECTOR*TRAIN_IMG_NUM);//总样本的特征向量
	vector<float>::iterator pos;
	pos = totaldescrip.begin();
	int ImageNum = 0;
	//读取文件地址
	String img_file = "C:\\Users\\tudejiang\\Desktop\\超声图像VOC格式标准数据集201911\\VOC2007\\JPEGImages\\*.jpg";
	//存储读取文件地址图片名
	vector<String> image_files;
	//存储图片函数
	glob(img_file, image_files);
	for (int i = 0; i < image_files.size(); i++)
	{
		cout << image_files[i] << endl;
		Mat src = imread(image_files[i]);
		cvtColor(src,src,COLOR_BGR2GRAY);
		hog.compute(src,descrip);//计算每张图像的HOG特征向量
		vector<float>::iterator iter;
		for (iter = descrip.begin(); iter != descrip.end(); iter++, pos++)
		{
			*pos = *iter;
		}
	}
	//制作训练样本
	float* buf2 = &totaldescrip[0];
	Mat data_mat(TRAIN_IMG_NUM, HOG_VECTOR, CV_32FC1, buf2);
	//制作训练标签
	Mat label_mat=Mat::ones(TRAIN_IMG_NUM,1,CV_32SC1);
	for (int k = 400; k < 800; k++)
	{
		label_mat.at<unsigned int>(k, 0) = 2;
	}
	//利用SVM进行训练生成XML文件
	Ptr<ml::SVM> svm = ml::SVM::create();
	svm->setTermCriteria(TermCriteria(TermCriteria::EPS, 1000, FLT_EPSILON));
	svm->setType(SVM::C_SVC);
	svm->setKernel(SVM::RBF);
	svm->setDegree(10.0);//高斯核的参数设置
	svm->setGamma(8.0);
	svm->setCoef0(1.0);
	svm->setC(10.0);
	svm->setNu(0.5);
	svm->setP(0.1);
	Ptr<TrainData> traind = TrainData::create(data_mat, ROW_SAMPLE, label_mat);
	svm->train(traind);
	svm->save("svm_image.xml");
	return 0;
}

2、给出图片预测

#include<iostream>
#include<opencv2/opencv.hpp>
#include<vector>
using namespace std;
using namespace cv;
using namespace cv::ml;
#define HOG_VECTOR 900
#define TRAIN_IMG_NUM 800
int main()
{
	//创建HOG描述子
	HOGDescriptor hog(Size(48, 48), Size(16, 16), Size(8, 8), Size(8, 8), 9,-1);
	//提取样本特征
	vector<float> descrip1;//每个检测窗口的特征向量
	Mat src1 = imread("renlianjiance.jpg");
	hog.compute(src1,descrip1);
	float* buf1 = &descrip1[0];
	Mat hog_data_mat(1,900,CV_32FC1,buf1);
	//利用SVM进行训练生成XML文件
	Ptr<ml::SVM> svm = ml::SVM::create();
	svm->load("svm_image.xml");
	float index = 0;
	index = svm->predict(hog_data_mat);
	cout << index << endl;
	return 0;
}