Ptr<FaceRecognizer> Trainer::load(const char* fname,int type)
{
Ptr<FaceRecognizer> model;
if(type==0)
model=createFisherFaceRecognizer();
else
model=createLBPHFaceRecognizer();
model->load(fname);
char lable_fname[MAX_PATH];
sprintf(lable_fname,"%s.labels",fname);
FILE* fp=fopen(lable_fname,"rt");
if(fp==0)
{
printf("Fail to load .labels file, please re-train the model\n");
return NULL;
}
int class_count=0;
fscanf(fp,"%d",&class_count);fgetc(fp);
names.clear();
char buf[256];
for(int i=0;i<class_count;++i)
{
fscanf(fp,"%s",buf);
names.push_back(buf);
}
fclose(fp);
return model;
}
void FacialFeatureRecognizerTest::compareSmileFrownSimilarFacesTest()
{
Ptr<FaceRecognizer> model = createFisherFaceRecognizer();
FacialFeatureRecognizer recognizer(model, 0, NULL, NULL, NULL);
QString trainingFile("/home/zane/Documents/COS301/training.xml");
recognizer.loadTrainingFromXML(trainingFile);
Mat face1 = imread("../../testFiles/FaceRec/barack_smile.jpg", CV_LOAD_IMAGE_UNCHANGED);
Mat face2 = imread("../../testFiles/FaceRec/barack_frown.jpg", CV_LOAD_IMAGE_UNCHANGED);
string faceCascade = "/home/zane/Documents/COS301/MainProject/testFiles/haarcascade_frontalface_alt2.xml";
Filter* faceDetect = new FaceDetectFilter(faceCascade);
Filter* preProc = new PreProcessingFilter(140, 150);
ImageData* data1 = new ImageData(face1, 0);
data1 = faceDetect->filter(data1);
data1 = preProc->filter(data1);
ImageData* data2 = new ImageData(face2, 0);
data2 = faceDetect->filter(data2);
data2 = preProc->filter(data2);
double expected = 5800;
double actual = recognizer.compareFaces(data1->faces[0], data2->faces[0]);
QVERIFY(actual <= expected);
}
void TrackFace::training()
{
model=createFisherFaceRecognizer();
model->train(images, labels);
model->save(fn_features);
}
void ofxFaceRecognizer::setup(int method_used, int _maxFaces, bool bAlreadySavedModel, string folderName) {
//eigen take much longer to load and longer to generate. also makes much larger yml file
string method_names[3] = {"eigen","fisher","lbph"};
// Create a FaceRecognizer and train it on the given images:
methodId = method_used;
methodName = method_names[method_used];
if(methodId == 0){
model = createEigenFaceRecognizer();
}
if(methodId == 1){
model = createFisherFaceRecognizer();
}
if(methodId == 2){
model = createLBPHFaceRecognizer();
}
//if(_maxFaces > 0){
maxFaces = _maxFaces;
//by default my training images should be 150x150 pixels
//will reset if you use loadTrainingImages()
image_width = 150;
image_height = 150;
//load in training images
loadTrainingImages(folderName,maxFaces);
string compiledDatabasePath = ofToDataPath("model_"+methodName+"_"+ofToString(maxFaces)+"_"+folderName+".yml");
if(bAlreadySavedModel){
cout<<"model .yml supposedly existing"<<endl;
model->load(compiledDatabasePath);
cout<<"loaded "<<maxFaces<<" faces with model "<<methodName<<endl;
}else{
cout<<"start training new model. this might take a very long time"<<endl;
cout<<"compiledDatabasePath "<<compiledDatabasePath<<endl;
cout<<"more so for fisher than eigen"<<endl;
model->train(allTrainingMats, allTrainingLabels);
model->save(ofToDataPath(compiledDatabasePath));
cout<<"trained and saved .yml with "<<maxFaces<<" faces with model "<<methodName<<endl;
}
// Quit if there are not enough images for this demo.
if(allTrainingMats.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);
}
}
DWORD WINAPI FaceRecog::FaceRecogThread()
{
vector<Mat> images;
vector<int> labels;
vector<string> names;
// Read in the data (fails if no valid input filename is given, but you'll get an error message):
try {
Read_CSV(CSVFileName, images, labels, names);
} catch (cv::Exception& e) {
cerr << "Error opening file \"" << CSVFileName << "\". Reason: " << e.msg << endl;
// nothing more we can do
exit(1);
}
// Get unique names
getUniqStrs(names, m_uniq_names);
m_imwidth = images[0].cols;
m_imheight = images[0].rows;
m_pmodel = createFisherFaceRecognizer();
m_pmodel->train(images, labels);
//m_pmodel->load("C:\\Users\\yuner\\Desktop\\Fisher_at.yml");
m_haar_cascade.load(HaarCascadeFileName);
//Capture a frame and process it in a loop
while (1)
{
processFrame();
Sleep(16);
}
return 0;
}
facerecognize::facerecognize()
{
this->model = createFisherFaceRecognizer(20);
//this->model = createLBPHFaceRecognizer(1, 5, 5, 5);
this->faceDetec = new facedetec();
this->loaded = false;
this->readcsved = false;
this->im_width = 92;
this->im_height = 112;
this->output_folder = "./face.xml";
mapint[1] = "WL";
mapint[2] = "JHZ";
mapint[3] = "XSY";
mapint[4] = "ZZC";
mapint[5] = "CZM";
mapint[6] = "XD";
mapint[7] = "YYF";
mapint[8] = "WYQ";
mapint[9] = "YY";
}
void Trainer::train(const char* saveto,int type)
{
Ptr<FaceRecognizer> model;
if(type==0)
model=createFisherFaceRecognizer();
else
model=createLBPHFaceRecognizer();
printf("\nbegin training...\n");
model->train(images, labels);
model->save(saveto);
char lable_fname[MAX_PATH];
sprintf(lable_fname,"%s.labels",saveto);
FILE* fp=fopen(lable_fname,"wt");
fprintf(fp,"%d\n",names.size());
for(int i=0,iLen=names.size();i<iLen;++i)
{
fprintf(fp,"%s\n",names[i].c_str());
}
fclose(fp);
printf("Trained model saved to \"%s\"\n",saveto);
}
int main( int argc, char** argv )
{
// Medida de tiempo inicial
clock_t t0, t_load, t_loop, t_loop0;
double dtime;
t0 = clock();
// Declaration of variables
CascadeClassifier face_cascade; // Cascade Classifier
Mat captureFrame, grayscaleFrame; // Captured and converted to gray Frames
double x_face_pos, y_face_pos, area_face; // Coordinates of the detected face
vector< Rect_<int> > faces; // Vector of faces
#ifdef RASPBERRY
RaspiCamCvCapture * captureDevice; // Video input
#else
CvCapture * captureDevice; // Video input
#endif
char sTmp[255];
#ifdef TRACE
sprintf(sTmp,"\n Directorio de ejecucion del programa: ");
trace (sTmp);
cout << get_ProgramDirectory();
#endif
#ifdef RECOGNITION
// Declaration of variables Face recognition
string people[MAX_PEOPLE]; // Each person of the model of face recognition
int im_width, im_height; // heigh, witdh of 1st images of the model of face recognition
int prediction_seuil; // Prediction limit
Ptr<FaceRecognizer> model; // Model of face recognitio
// Prediction limit depending on the device
#ifdef EIGENFACES
prediction_seuil = 10;
#else
prediction_seuil = 1000;
#endif
// Model of face recognition
#ifdef EIGENFACES
model = createEigenFaceRecognizer();
#else
model = createFisherFaceRecognizer();
#endif
// Read measures file
read_measures_file(im_width, im_height, '=');
// Read people file
read_people_file(people, '(', ')', '=');
// Load model
load_model_recognition(model);
t_load = clock();
#ifdef DEBUG
dtime = difftime(t_load,t0);
sprintf(sTmp,"\n (Face Tracking) tiempo de carga del modelo = ");
debug(sTmp);
cout << print_time(dtime);
#endif
#endif
// Video input depending on the device
#ifdef RASPBERRY
captureDevice = raspiCamCvCreateCameraCapture(0); // Index doesn't really matter
#else
captureDevice = cvCreateCameraCapture(0);
#endif
// Load of Haar Cascade
if (!load_haar_cascade(face_cascade)) { return -1;}
// Create new window
SHOW cvNamedWindow("Face tracking", 1);
do {
t_loop0 = clock();
#ifdef RASPBERRY
IplImage* image = raspiCamCvQueryFrame(captureDevice); // Get images from the video input
#else
IplImage* image = cvQueryFrame(captureDevice); // Get images from the video input
#endif
captureFrame = cvarrToMat(image); // Convert images to Mat
cvtColor(captureFrame, grayscaleFrame, CV_RGB2GRAY); // Convert the image to gray scale
// Detection and Face Recognition
face_detection(face_cascade, grayscaleFrame, captureFrame, &faces, x_face_pos, y_face_pos, area_face);
#ifdef RECOGNITION
// Detection and Face Recognition
face_recognition(people, grayscaleFrame, captureFrame, &faces, im_width, im_height, model,
prediction_seuil, x_face_pos, y_face_pos, area_face);
#endif
// Display results
#ifdef SHOW
imshow("Face tracking", captureFrame);
#endif
t_loop = clock();
#ifdef DEBUG
dtime = difftime(t_loop,t_loop0);
sprintf(sTmp,"\n (Face Tracking) tiempo del bucle del reconocimiento de cara = ");
debug(sTmp);
cout << print_time(dtime);
#endif
} while(cvWaitKey(10) < 0);
// Destroy window
#ifdef SHOW
cvDestroyWindow("Face tracking");
#endif
#ifdef TRACE
trace ("\n");
#endif
#ifdef RASPBERRY
raspiCamCvReleaseCapture(&captureDevice);
#else
cvReleaseCapture(&captureDevice);
#endif
return 0;
}
GenderRecognizer::GenderRecognizer(void)
{
model = createFisherFaceRecognizer();
model->load("data\\gender_classifier_25imgs.xml");
}
ICLASS_API FaceRecognizer* __stdcall Create_FisherFaceRecognizer(int num_components, double threshold)
{
Ptr<FaceRecognizer> r = createFisherFaceRecognizer(num_components, threshold);
r.addref();
return r;
}
Ptr<FaceRecognizer> gender_detection(string fn_csv)
{
string output_folder;
//string fn_csv = string("/Users/xueqianjiang/Desktop/male.txt");
// These vectors hold the images and corresponding labels.
vector<Mat> images;
vector<int> labels;
// Read in the data. This can fail if no valid
// input filename is given.
try {
read_csv(fn_csv, images, labels);
} catch (Exception& e) {
cerr << "Error opening file \"" << fn_csv << "\". Reason: " << e.msg << endl;
// nothing more we can do
exit(1);
}
if(images.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);
}
// this part will add in the EigenFaceRecognizer in order to in reduce the dimension further
// int height = images[0].rows;
//PCA model
//Ptr<FaceRecognizer> model = createEigenFaceRecognizer();
//train the PCA model and without loss of information
//model->train(images, labels);
//save the results of the train
// model->save("eigenface.yml");
//take out the feature values from the eigenfaces and then rank them from largest to smallest
//Mat eigenvalues = model->getMat("eigenvalues");
//take out the feature values from the model，eigenvectures and the eigenvalues are in the same order
//Mat W = model->getMat("eigenvectors");
//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/所有特征值总和>0.97"；其次，121=11^2，可以将结果表示成一个11*11的2维图像方阵，交给fisherface去计算
// int xth = 121;
//after PCA
//vector<Mat> reduceDemensionimages;
//choose the first xth eigenvalues and get rid of the rest
// Mat evs = Mat(W, Range::all(), Range(0, xth));
//for(int i=0;i<images.size();i++)
//{
// Mat mean= model->getMat("mean");
//subspaceProjection
//Mat projection = subspaceProject(W, mean, images[i].reshape(1,1));
//reduced dimensionality and save them
//reduceDemensionimages.push_back(projection.reshape(1,sqrt(xth)));
// }
// Quit if there are not enough images for this demo.
// Ptr<FaceRecognizer> fishermodel = createFisherFaceRecognizer();
//fishermodel->train(reduceDemensionimages,labels);
Ptr<FaceRecognizer> model = createFisherFaceRecognizer();
model->train(images, labels);
return model;
}
