int main() {
std::vector<ContourWithData> allContoursWithData;
std::vector<ContourWithData> validContoursWithData;
cv::Mat matClassificationFloats;
cv::FileStorage fsClassifications("classifications.xml", cv::FileStorage::READ);
if (fsClassifications.isOpened() == false) {
std::cout << "error, unable to open training classifications file, exiting program\n\n";
return(0);
}
fsClassifications["classifications"] >> matClassificationFloats;
fsClassifications.release();
cv::Mat matTrainingImages;
cv::FileStorage fsTrainingImages("images.xml", cv::FileStorage::READ);
if (fsTrainingImages.isOpened() == false) {
std::cout << "error, unable to open training images file, exiting program\n\n";
return(0);
}
fsTrainingImages["images"] >> matTrainingImages;
fsTrainingImages.release();
cv::Ptr<cv::ml::KNearest> kNearest = cv::ml::KNearest::create();
cv::Ptr<cv::ml::TrainData> trainingData = cv::ml::TrainData::create(matTrainingImages, cv::ml::SampleTypes::ROW_SAMPLE, matClassificationFloats);
kNearest->setIsClassifier(true);
kNearest->setAlgorithmType(cv::ml::KNearest::Types::BRUTE_FORCE);
kNearest->setDefaultK(101);
cv::Mat matResults(0, 0, CV_32F);
kNearest->train(trainingData);
cv::Mat matTestingNumbers = cv::imread("test_numbers.png");
if (matTestingNumbers.empty()) {
std::cout << "error: image not read from file\n\n";
return(0);
}
cv::Mat matGrayscale;
cv::Mat matBlurred;
cv::Mat matThresh;
cv::Mat matThreshCopy;
cv::cvtColor(matTestingNumbers, matGrayscale, CV_BGR2GRAY);
threshold(matGrayscale, matThresh, 10, 255, CV_THRESH_BINARY_INV);
matThreshCopy = matThresh.clone();
std::vector<std::vector<cv::Point> > ptContours;
std::vector<cv::Vec4i> v4iHierarchy;
cv::findContours(matThreshCopy,
ptContours,
v4iHierarchy,
CV_RETR_EXTERNAL,
CV_CHAIN_APPROX_TC89_KCOS);
for (int i = 0; i < ptContours.size(); i++) {
ContourWithData contourWithData;
contourWithData.ptContour = ptContours[i];
contourWithData.boundingRect = cv::boundingRect(contourWithData.ptContour);
contourWithData.fltArea = cv::contourArea(contourWithData.ptContour);
allContoursWithData.push_back(contourWithData);
}
for (int i = 0; i < allContoursWithData.size(); i++) {
if (allContoursWithData[i].checkIfContourIsValid()) {
validContoursWithData.push_back(allContoursWithData[i]);
}
}
std::sort(validContoursWithData.begin(), validContoursWithData.end(), ContourWithData::sortByBoundingRectXPosition);
std::string strFinalString;
for (int i = 0; i < validContoursWithData.size(); i++) {
cv::rectangle(matTestingNumbers,
validContoursWithData[i].boundingRect,
cv::Scalar(0, 255, 0),
2);
cv::Mat matROI = matThresh(validContoursWithData[i].boundingRect);
cv::Mat matROIResized;
cv::resize(matROI, matROIResized, cv::Size(RESIZED_IMAGE_WIDTH, RESIZED_IMAGE_HEIGHT));
cv::Mat matROIFloat;
matROIResized.convertTo(matROIFloat, CV_32FC1);
float fltCurrentChar = kNearest->findNearest(matROIFloat.reshape(1,1), kNearest->getDefaultK(), matResults);
strFinalString = strFinalString + char(int(fltCurrentChar));
}
cv::Mat string_box(100,500,CV_8UC3, cv::Scalar::all(0));
int baseLine = 0;
cv::Size string_size = cv::getTextSize(strFinalString, CV_FONT_HERSHEY_DUPLEX, 1, 2, &baseLine);
baseLine += 2;
cv::Point textOrg((string_box.cols - string_size.width) / 2, (string_box.rows + string_size.height) / 2);
cv::putText(string_box, strFinalString, textOrg, CV_FONT_HERSHEY_DUPLEX, 1, cv::Scalar::all(255), 2, 8);
cv::imshow("result", string_box);
cv::imshow("matTestingNumbers", matTestingNumbers);
cv::waitKey(0);
return(0);
}
void PlateRecognizer::Process()
{
knn_ = cv::Algorithm::load < cv::ml::KNearest>( "C:\\Users\\Jonathan\\Documents\\ESGI\\5A\\machine_learning\\datasets\\savedDatas"); //chargement des données entrainées
for ( std::vector<Image>::iterator it = images_.begin(); it != images_.end(); ++it )//pour chacune des images du dossier
{
cv::Mat image1 = ( *it ).GetImage();//on stock dans une variable temp l'image en cours
std::vector<Plate> tmpplates = std::vector<Plate>();
if ( Detection( image1, tmpplates ) )//detections des plaques temporaires
{
( *it ).SetPlates( tmpplates );
cv::Mat result;
image1.copyTo( result );
if ( showResult )//si jamais ce booleen est a true on affiche les rectangle autours des plaques détectées
{
for ( std::vector<Plate>::const_iterator pit = ( *it ).GetPlates().cbegin(); pit != ( *it ).GetPlates().cend(); ++pit )
{
cv::Point2f rect_points[4];
( *pit ).GetRect().points( rect_points );
for ( int j = 0; j < 4; j++ )
line( result, rect_points[j], rect_points[( j + 1 ) % 4], cv::Scalar( 255, 100, 0 ), 5, 8 );
}
cv::imshow( "PlateRecognizer", result );
cv::waitKey();
}
for ( std::vector<Plate>::iterator pit = ( *it ).GetPlates().begin(); pit != ( *it ).GetPlates().end(); ++pit )//pour chacunes des plaques detectées
{
CharactersDetection( *pit );//on detecte les caractere et on rempli le vector chars de la Plate courante
std::cout << "number of chars detected = " << ( *pit ).GetChars().size() << std::endl;
}
const int size = 20; //la taille de chaque caractere (a etre redimensionné)
const int tolerence = 60; //la tolerance : chaque pixel est entre 0 et 255 pour du nb on choisis une valeur tolerance pour dire : plus petit que la tolerance = 0 sinon 1
for ( std::vector<Plate>::const_iterator pit = ( *it ).GetPlates().cbegin(); pit != ( *it ).GetPlates().cend(); ++pit )//pour chacune des plaques
{
std::vector<char> letters = std::vector<char>();
for ( std::vector<Chars>::const_iterator cit = ( *pit ).GetChars().cbegin(); cit != ( *pit ).GetChars().cend(); ++cit )//pour chacuns des carateres de la plaque
{
cv::Mat resized;
cv::resize( ( *cit ).m_char, resized, cv::Size( size, size ) );//on resize en size*size (meme valeur que pour le dataset)
cv::Mat matResults( 1, 20 * 20, CV_32F );
for ( int i = 0; i < 20 * 20; ++i )
matResults.at<float>( cv::Point( i, 0 ) ) = ( (int) ( resized.data[i] ) > tolerence ) ? 0 : 1; //on change la valeur des pixels en 0 ou 1 (comme pour le dataset)
int f = knn_->predict( matResults );//resultat du predict avec le caractere courant
std::cout << "result: " << chars[f] << std::endl;//affichage du caractere trouvé par le KNN
if (f != -1)//-1 pour le bruit, pas encore geré
letters.push_back( chars[f] );//remplissage de la chaine de caractere correspondante a la plaque
if ( showResult )
{
cv::imshow( "Chars", resized );
cv::waitKey();
}
}
std::cout << "Found for this image : ";
for ( auto it = letters.begin() ; it != letters.end() ; ++it )
std::cout << *it;//affichage de la chaine de caractere trouvé correspondante a la plaque
std::cout << std::endl;
}
}
else
std::cout << "No Plate found in this picture" << std::endl;
}
}
