上一篇中我們對訓練資料做了一些預處理，檢測出人臉並儲存在\pic\color\x資料夾下（x=1,2,3,...類別號），本文做訓練和識別。為了識別，首先將人臉訓練資料 轉為灰度、對齊、歸一化，再放入分類器（EigenFaceRecognizer）,最後用訓練出的model進行predict。

-----------------------------------------

環境：vs2010+opencv 2.4.6.0

特徵：eigenface

Input：一個人臉資料庫，15個人，每人20個樣本（左右）。

Output：人臉檢測，並識別出每張檢測到的人臉。

-----------------------------------------

1. 為訓練資料預處理（ 轉為灰度、對齊、歸一化 ）

• 轉為灰度和對齊是後面做訓練時EigenFaceRecognizer的要求；
• 歸一化是防止光照帶來的影響

在上一篇的 2.2 Prehelper.cpp檔案中加入函式

void resizeandtogray(char* dir,int k, vector<Mat> &images, vector<int> &labels,
vector<Mat> &testimages, vector<int> &testlabels);

void resizeandtogray(char* dir,int K, vector<Mat> &images, vector<int> &labels,
	vector<Mat> &testimages, vector<int> &testlabels)
{
	IplImage* standard = cvLoadImage("D:\\privacy\\picture\\photo\\2.jpg",CV_LOAD_IMAGE_GRAYSCALE);
	string cur_dir;
	char id[5];
	int i,j;
	for(int i=1; i<=K; i++)
	{
		cur_dir = dir;
		cur_dir.append("gray\\");	
		_itoa(i,id,10);
		cur_dir.append(id);
		const char* dd = cur_dir.c_str();
		CStatDir statdir;
		if (!statdir.SetInitDir(dd))
		{
			puts("Dir not exist");
			return;
		}
		cout<<"Processing samples in Class "<<i<<endl;
		vector<char*>file_vec = statdir.BeginBrowseFilenames("*.*");
		for (j=0;j<file_vec.size();j++)
		{
			IplImage* cur_img = cvLoadImage(file_vec[j],CV_LOAD_IMAGE_GRAYSCALE);
			cvResize(cur_img,standard,CV_INTER_AREA);
			Mat cur_mat = cvarrToMat(standard,true),des_mat;
			cv::normalize(cur_mat,des_mat,0, 255, NORM_MINMAX, CV_8UC1);
			cvSaveImage(file_vec[j],cvCloneImage(&(IplImage) des_mat));
			if(j!=file_vec.size())
			{
					images.push_back(des_mat);
					labels.push_back(i);
			}
			else
			{
				testimages.push_back(des_mat);
				testlabels.push_back(i);
			}
		}
		cout<<file_vec.size()<<" images."<<endl;
	}
}
 

並在main中呼叫：

int main( )
{
	CvCapture* capture = 0;
	Mat frame, frameCopy, image;
	string inputName;	
	int mode;

	char dir[256] = "D:\\Courses\\CV\\Face_recognition\\pic\\"; 
	//preprocess_trainingdata(dir,K); //face_detection and extract to file
	vector<Mat> images,testimages;
	vector<int> labels,testlabels;
	resizeandtogray(dir,K,images,labels,testimages,testlabels); //togray, normalize and resize
	
	system("pause");
	return 0;
}
 

2. 訓練

有了vector<Mat> images,testimages;vector<int> labels,testlabels; 可以開始訓練了，我們採用EigenFaceRecognizer建模。

在Prehelper.cpp中加入函式

Ptr<FaceRecognizer> Recognition(vector<Mat> images, vector<int> labels,vector<Mat> testimages, vector<int> testlabels)；

Ptr<FaceRecognizer> Recognition(vector<Mat> images, vector<int> labels,
	vector<Mat> testimages, vector<int> testlabels)
{
	Ptr<FaceRecognizer> model = createEigenFaceRecognizer(10);//10 Principal components
	cout<<"train"<<endl;
	model->train(images,labels);
	int i,acc=0,predict_l;
	for (i=0;i<testimages.size();i++)
	{
		predict_l = model->predict(testimages[i]);
		if(predict_l != testlabels[i])
		{
			cout<<"An error in recognition: sample "<<i+1<<", predict "<<
				predict_l<<", groundtruth "<<testlabels[i]<<endl;
			imshow("error 1",testimages[i]);
			waitKey();
		}
		else
			acc++;
	}
	cout<<"Recognition Rate: "<<acc*1.0/testimages.size()<<endl;
	return model;
}

主函式改為：

int main( )
{
	CvCapture* capture = 0;
	Mat frame, frameCopy, image;
	string inputName;	
	int mode;

	char dir[256] = "D:\\Courses\\CV\\Face_recognition\\pic\\"; 
	//preprocess_trainingdata(dir,K); //face_detection and extract to file
	vector<Mat> images,testimages;
	vector<int> labels,testlabels;
	//togray, normalize and resize; load to images,labels,testimages,testlabels
	resizeandtogray(dir,K,images,labels,testimages,testlabels); 
	//recognition
	Ptr<FaceRecognizer> model = Recognition(images,labels,testimages,testlabels);
	char* dirmodel = new char [256];
	strcpy(dirmodel,dir); strcat(dirmodel,"model.out");
	FILE* f = fopen(dirmodel,"w");
	fwrite(model,sizeof(model),1,f);
	system("pause");
	return 0;
}

最終結果：一個錯分樣本，正確率93.3%

關於Computer Vision更多的學習資料將繼續更新，敬請關注本部落格和新浪微博Rachel Zhang。