SVM+HOG特征训练分类器

  1 #include <iostream>    
  2 #include <fstream>    
  3 #include <string>    
  4 #include <vector> 
  5 #include <stdlib.h>
  6 #include <tchar.h>
  7 #include <windows.h>
  8    
  9 #include <opencv2/core/core.hpp>
 10 #include <opencv2/ml/ml.hpp>
 11 #include <opencv2/highgui/highgui.hpp>
 12 #include <opencv2/objdetect/objdetect.hpp>
 13 #include <opencv2/imgproc/imgproc.hpp>
 14 
 15 using namespace cv;    
 16 using namespace std;    
 17    
 18 int getfilepath(string txtpath,vector<string>& img_path, vector<int>& roi_sample_class, vector<vector<Rect > >& roi_sample_rect);  
 19   
 20 int main(int argc, char** argv)  {  
 21     //winSize窗口大小
 22     int ImgWidht = 48;  
 23     int ImgHeight = 48; 
 24     
 25     //Sample总数
 26     int num_sample_roi=0;
 27 
 28     //图片路径，每张图片中的sample类别和每个sample的rect参数
 29     vector<string> img_path;
 30     vector<int> roi_sample_class;
 31     vector<vector<Rect > > roi_sample_rect;
 32     
 33     //标记文件txt名称和路径
 34     string filepath = "E:/svm/";
 35     string txtpath = string(filepath) + "TrainingData.txt";
 36     
 37     //获得图片路径，sample类别，每个sample的rect参数，返回Sample总数(ROI总数)
 38     num_sample_roi=getfilepath(txtpath, img_path, roi_sample_class, roi_sample_rect);
 39     
 40     //测试img_path, roi_sample_class, roi_sample_rect
 41     //cout << "img_path[0]= " << img_path[0] << "\n img_path[50]= " << img_path[50] << endl;
 42     //cout << "roi_sample_class[0]= " << roi_sample_class[0] << "\n roi_sample_class[150]= " << roi_sample_class[150] << endl;
 43     //cout << "roi_sample_rect[0][0]= " << roi_sample_rect[0][0] << endl;
 44     //system("Pause");
 45 
 46     //HOG特征矩阵，sample类别矩阵
 47     //cout << "num_sample_roi= " << num_sample_roi << endl;
 48     //system("Pause");
 49     Mat sample_feature_mat(num_sample_roi, 900, CV_32FC1);//900=(win_height/8-1)*(win_width/8-1)*(2*2)*9;
 50     Mat sample_class_mat(num_sample_roi, 1, CV_32SC1);    //样本类别       
 51 
 52     //原图片和训练图片
 53     Mat orig_img;    
 54     Mat train_img; //= Mat::zeros(ImgWidht, ImgHeight, CV_8UC3);//需要分析的图片 
 55 
 56     //sample指示子
 57     unsigned long n_sample = 0;
 58     
 59     //对图片循环 
 60     for( string::size_type i = 0; i != img_path.size(); i++ )  
 61     {    
 62         orig_img = imread(img_path[i].c_str(), 1); 
 63         if(orig_img.empty()){
 64             cout<<"Can not load the image: "<<img_path[i]<<endl;
 65             continue;
 66         }
 67         
 68         //端口的提示信息
 69         cout<<"***processing***"<<img_path[i].c_str()<<endl;        
 70         
 71         //每个sample都要计算hog特征
 72         for (size_t j = 0; j != roi_sample_rect[i].size(); j++){
 73             //取ROI，归一化
 74             Mat handle_src=orig_img(roi_sample_rect[i][j]);
 75             resize(handle_src, train_img, Size(ImgWidht, ImgHeight));
 76             
 77             //申明描述子，每个参数的含义见笔记
 78             HOGDescriptor hog(Size(ImgWidht,ImgHeight),Size(16,16),Size(8,8),Size(8,8), 9);
 79             
 80             //描述子申请内存并计算
 81             vector<float> descriptors; 
 82             hog.compute(train_img, descriptors);
 83             
 84             //为当前sample的所有hog descriptor申请内存
 85             //sample_feature_mat[n_sample].resize(descriptors.size(),CV_32FC1);
 86             
 87             //输出hog特征个数
 88             //cout<<"HOG dims: "<<descriptors.size()<<endl; 
 89             
 90             //每个sample的hog特征个数    
 91             for (vector<float>::size_type k = 0; k != descriptors.size(); k++)
 92                 sample_feature_mat.at<float>(n_sample, k) = descriptors[k]; 
 93             
 94             //int num_class=i/100;
 95             sample_class_mat.at<int>(n_sample, 0) =  roi_sample_class[i];    
 96             cout<<"***end processing***"<<img_path[i].c_str()<<" "<<roi_sample_class[i]<<endl;
 97             
 98             n_sample++;
 99         }
100         
101     }    
102     
103     //SVM参数
104     Ptr<ml::SVM> svm = ml::SVM::create();
105     svm->setType(ml::SVM::C_SVC);
106     svm->setKernel(ml::SVM::RBF);
107     svm->setTermCriteria(TermCriteria(CV_TERMCRIT_EPS, 1000, FLT_EPSILON));
108     
109     //SVM训练    
110     svm->train(sample_feature_mat, ml::ROW_SAMPLE, sample_class_mat);
111     svm->save( filepath+"SVM_DATA.xml" );      
112     
113     system("Pause");
114     return 0;    
115 }  
116 
117 
118 int getfilepath(string txtpath,vector<string>& img_path, vector<int>& roi_sample_class, vector<vector<Rect > >& roi_sample_rect){
119     int nLine = 0;    //图片计数器，每个类别100张图片。
120     int numroi=0;            //sample计数器，总的roi个数
121     
122     ifstream svm_data( txtpath );
123     char output[1000];
124     
125     while( !svm_data.eof() ) {    
126         svm_data >> output;//是不是>>遇见空格会自动截断？
127         string s0= string(output);
128         if( s0.length()>10 ){
129             if( nLine < 100 ){    
130                 roi_sample_class.push_back(0); 
131                 
132                 size_t bufname=s0.find(" ");
133                 string bufname1 = s0.substr(0, bufname);
134                 img_path.push_back( bufname1 ); 
135                 
136                 vector<Rect> obj_list;
137                 svm_data >> output;
138                 int obj_num=atoi(output);
139                 
140                 numroi=numroi+obj_num;
141                 int obj[4];
142                 for(int i=0; i<obj_num;i++)
143                 {
144                     for(int j=0;j<4;j++)
145                     {
146                         svm_data >> output;
147                         int p=atoi(output);
148                         obj[j]=p;
149                     }
150                     Rect tmp_obj=Rect(obj[0],obj[1],obj[2],obj[3]);
151                     obj_list.push_back(tmp_obj);
152                 }
153                 roi_sample_rect.push_back(obj_list);
154             }    
155             else if( nLine < 200 ){    
156                 roi_sample_class.push_back(1); 
157                 size_t bufname=s0.find(" ");
158                 string bufname1 = s0.substr(0, bufname);
159                 img_path.push_back( bufname1 ); 
160                 
161                 vector<Rect> obj_list;
162                 svm_data>>output;
163                 int obj_num=atoi(output);
164                 
165                 numroi=numroi+obj_num;
166                 int obj[4];
167                 for(int i=0; i<obj_num;i++)
168                 {
169                     for(int j=0;j<4;j++)
170                     {
171                         svm_data >> output;
172                         int p=atoi(output);
173                         obj[j]=p;
174                     }
175                     Rect tmp_obj=Rect(obj[0],obj[1],obj[2],obj[3]);
176                     obj_list.push_back(tmp_obj);
177                 }
178                 roi_sample_rect.push_back(obj_list);
179             }
180             else if( nLine < 300 ){    
181                 roi_sample_class.push_back(2); 
182                 size_t bufname=s0.find(" ");
183                 string bufname1 = s0.substr(0, bufname);
184                 img_path.push_back( bufname1 ); 
185                 
186                 vector<Rect> obj_list;
187                 svm_data>>output;
188                 int obj_num=atoi(output);
189                 
190                 numroi=numroi+obj_num;
191                 int obj[4];
192                 for(int i=0; i<obj_num;i++)
193                 {
194                     for(int j=0;j<4;j++)
195                     {
196                         svm_data >> output;
197                         int p=atoi(output);
198                         obj[j]=p;
199                     }
200                     Rect tmp_obj=Rect(obj[0],obj[1],obj[2],obj[3]);
201                     obj_list.push_back(tmp_obj);
202                 }
203                 roi_sample_rect.push_back(obj_list);
204             }
205             else{//(nLine < 400)
206                 roi_sample_class.push_back(3); 
207                 size_t bufname=s0.find(" ");
208                 string bufname1 = s0.substr(0, bufname);
209                 img_path.push_back( bufname1 ); 
210                 
211                 vector<Rect> obj_list;
212                 svm_data>>output;
213                 int obj_num=atoi(output);
214                 
215                 numroi=numroi+obj_num;
216                 int obj[4];
217                 for(int i=0; i<obj_num;i++)
218                 {
219                     for(int j=0;j<4;j++)
220                     {
221                         svm_data >> output;
222                         int p=atoi(output);
223                         obj[j]=p;
224                     }
225                     Rect tmp_obj=Rect(obj[0],obj[1],obj[2],obj[3]);
226                     obj_list.push_back(tmp_obj);
227                 }
228                 roi_sample_rect.push_back(obj_list);
229             }
230             nLine ++;        //计数
231         }    
232     } 
233     
234     svm_data.close();
235     //cout << "numroi= " << numroi<< endl;
236     return numroi;
237 }