* change to train Chinese model.

include/easypr/config.h

@@ -28,11 +28,11 @@ static const float kSvmPercentage   = 0.7f;
 
 static const int   kCharacterInput  = 120;
 static const int   kChineseInput    = 440;
-static const int   kAnnInput = kCharacterInput;
+static const int   kAnnInput = kChineseInput;
 
 static const int   kCharacterSize = 10;
 static const int   kChineseSize = 20;
-static const int   kPredictSize = kCharacterSize;
+static const int   kPredictSize = kChineseSize;
 
 static const int   kNeurons       = 40;
 

include/easypr/core/feature.h

@@ -37,6 +37,10 @@ void getLBPFeatures(const cv::Mat& image, cv::Mat& features);
 //! get character feature
 cv::Mat charFeatures(cv::Mat in, int sizeData);
 
+//! get character feature
+cv::Mat charFeatures2(cv::Mat in, int sizeData);
+
+//! LBP feature + Histom feature
 void getLBPplusHistFeatures(const cv::Mat& image, cv::Mat& features);
 
 } /*! \namespace easypr*/

src/core/feature.cpp

@@ -92,6 +92,8 @@ void getLBPFeatures(const Mat& image, Mat& features) {
 }
 
 
+
+
 Mat charFeatures(Mat in, int sizeData) {
   const int VERTICAL = 0;
   const int HORIZONTAL = 1;
@@ -139,6 +141,53 @@ Mat charFeatures(Mat in, int sizeData) {
 }
 
 
+Mat charFeatures2(Mat in, int sizeData) {
+  const int VERTICAL = 0;
+  const int HORIZONTAL = 1;
+
+  // cut the cetner, will afect 5% perices.
+  Rect _rect = GetCenterRect(in);
+  Mat tmpIn = CutTheRect(in, _rect);
+  //Mat tmpIn = in.clone();
+
+  // Low data feature
+  Mat lowData;
+  resize(tmpIn, lowData, Size(sizeData, sizeData));
+
+  // Histogram features
+  Mat vhist = ProjectedHistogram(lowData, VERTICAL);
+  Mat hhist = ProjectedHistogram(lowData, HORIZONTAL);
+
+  // Last 10 is the number of moments components
+  int numCols = vhist.cols + hhist.cols + lowData.cols * lowData.cols;
+
+  Mat out = Mat::zeros(1, numCols, CV_32F);
+  // Asign values to
+
+  // feature,ANN的样本特征为水平、垂直直方图和低分辨率图像所组成的矢量
+
+  int j = 0;
+  for (int i = 0; i < vhist.cols; i++) {
+    out.at<float>(j) = vhist.at<float>(i);
+    j++;
+  }
+  for (int i = 0; i < hhist.cols; i++) {
+    out.at<float>(j) = hhist.at<float>(i);
+    j++;
+  }
+  for (int x = 0; x < lowData.cols; x++) {
+    for (int y = 0; y < lowData.rows; y++) {
+      out.at<float>(j) += (float)lowData.at <unsigned char>(x, y);
+      j++;
+    }
+  }
+
+  //std::cout << out << std::endl;
+
+  return out;
+}
+
+
 void getLBPplusHistFeatures(const Mat& image, Mat& features) {
   // TODO
   Mat grayImage;

src/train/ann_train.cpp

@@ -46,7 +46,7 @@ void AnnTrain::train() {
   int first_hidden_neurons = int(std::sqrt((m + 2) * N) + 2 * std::sqrt(N / (m + 2)));
   int second_hidden_neurons = int(m * std::sqrt(N / (m + 2)));
 
-  bool useTLFN = false;
+  bool useTLFN = true;
   if (!useTLFN) {
     layers.create(1, 3, CV_32SC1);
     layers.at<int>(0) = input_number;
@@ -68,29 +68,28 @@ void AnnTrain::train() {
   ann_->setLayerSizes(layers);
   ann_->setActivationFunction(cv::ml::ANN_MLP::SIGMOID_SYM, 1, 1);
   ann_->setTrainMethod(cv::ml::ANN_MLP::TrainingMethods::BACKPROP);
-  ann_->setTermCriteria(cvTermCriteria(CV_TERMCRIT_ITER, 10000, 0.001));
+  ann_->setTermCriteria(cvTermCriteria(CV_TERMCRIT_ITER, 30000, 0.0001));
   ann_->setBackpropWeightScale(0.1);
   ann_->setBackpropMomentumScale(0.1);
 
   //using raw data or raw + synthic data.
   //auto traindata = tdata();
-  auto traindata = sdata(100);
+  auto traindata = sdata(350);
 
   std::cout << "Training ANN model, please wait..." << std::endl;
   long start = utils::getTimestamp();
   ann_->train(traindata);
   long end = utils::getTimestamp();
-  std::cout << "Training done. Time elapse: " << (end - start) / (1000 * 60) << "minute"
-            << std::endl;
-
   ann_->save(ann_xml_);
-  std::cout << "Your ANN Model was saved to " << ann_xml_ << std::endl;
-
+ 
   test();
+  std::cout << "Your ANN Model was saved to " << ann_xml_ << std::endl;
+  std::cout << "Training done. Time elapse: " << (end - start) / (1000 * 60) << "minute"
+    << std::endl;
 }
 
 std::pair<std::string, std::string> AnnTrain::identifyChinese(cv::Mat input) {
-  cv::Mat feature = charFeatures(input, kPredictSize);
+  cv::Mat feature = charFeatures2(input, kPredictSize);
   float maxVal = -2;
   int result = -1;
 
@@ -116,7 +115,7 @@ std::pair<std::string, std::string> AnnTrain::identifyChinese(cv::Mat input) {
 
 
 std::pair<std::string, std::string> AnnTrain::identify(cv::Mat input) {
-  cv::Mat feature = charFeatures(input, kPredictSize);
+  cv::Mat feature = charFeatures2(input, kPredictSize);
   float maxVal = -2;
   int result = -1;
 
@@ -216,8 +215,8 @@ cv::Mat getSyntheticImage(const Mat& image) {
   Mat result = image.clone();
 
   if (rand_type % 2 == 0) {
-    int ran_x = rand() % 7 - 3;
-    int ran_y = rand() % 7 - 3;
+    int ran_x = rand() % 5 - 2;
+    int ran_y = rand() % 5 - 2;
 
     result = translateImg(result, ran_x, ran_y);
   }
@@ -276,7 +275,7 @@ cv::Ptr<cv::ml::TrainData> AnnTrain::sdata(size_t number_for_count) {
     fprintf(stdout, ">> Characters count: %d \n", matVec.size());
 
     for (auto img : matVec) {
-      auto fps = charFeatures(img, kPredictSize);
+      auto fps = charFeatures2(img, kPredictSize);
 
       samples.push_back(fps);
       labels.push_back(i);
@@ -319,7 +318,7 @@ cv::Ptr<cv::ml::TrainData> AnnTrain::tdata() {
     auto chars_files = utils::getFiles(sub_folder);
     for (auto file : chars_files) {
       auto img = cv::imread(file, 0);  // a grayscale image
-      auto fps = charFeatures(img, kPredictSize);
+      auto fps = charFeatures2(img, kPredictSize);
 
       samples.push_back(fps);
       labels.push_back(i);