no commit message

res/train/svm准确率提高历史.txt

@@ -1,23 +0,0 @@
-EasyPR 1.0:
-svm liner
-precise: 0.786207
-recall: 0.844444
-Fsocre: 0.814286
-
-EasyPR 1.1:
-svm liner
-precise: 0.96798
-recall: 0.933492
-Fsocre: 0.950423
-
-EasyPR 1.1:
-svm liner:
-precise: 0.968974
-recall: 0.964371
-Fsocre: 0.966667
-
-EasyPR 1.1:
-svm rbf(histom feature):
-precise: 0.997596
-recall: 0.985748
-Fsocre: 0.991637

res/train/训练说明.txt

@@ -1,4 +1,4 @@
 要训练时，注意看src/train下的两个cpp文件
 
-数据在data文件夹下，svm对应Plate_detect，ann对应chars_recognise。要先解压，然后放置到正确的路径下，或者
+数据在data文件夹下，svm对应plate_detect_svm，ann对应chars_recognise。要先解压，然后放置到正确的路径下，或者
 你可以修改cpp文件中的文件夹路径。

src/main/java/com/yuxue/easypr/core/CharsIdentify.java

@@ -89,7 +89,7 @@ public class CharsIdentify {
 
     public void loadModel(String s) {
         this.ann.clear();
-        ann=ANN_MLP.loadANN_MLP(s, "ann"); // 加载ann配置文件
+        ann=ANN_MLP.loadANN_MLP(s, "ann"); // 加载ann配置文件  图像转文字的训练库文件
     }
 
 
@@ -118,7 +118,7 @@ public class CharsIdentify {
         int result = -1;
         Mat output = new Mat(1, numAll, CV_32FC1);
 
-        ann.predict(f, output, 0);
+        ann.predict(f, output, 0);  // 预测结果
 
         int ann_min = (!isChinses) ? ((isSpeci) ? 10 : 0) : numCharacter;
         int ann_max = (!isChinses) ? numCharacter : numAll;
@@ -133,7 +133,6 @@ public class CharsIdentify {
                 result = j;
             }
         }
-
         return result;
     }
 

src/main/java/com/yuxue/easypr/core/CharsRecognise.java

@@ -40,7 +40,7 @@ public class CharsRecognise {
         if (0 == result) {
             for (int j = 0; j < matVec.size(); j++) {
                 Mat charMat = matVec.get(j);
-                // 默认首个字符块是中文字符
+                // 默认首个字符块是中文字符   第二个字符块是字母
                 String charcater = charsIdentify.charsIdentify(charMat, (0 == j), (1 == j));
                 plateIdentify = plateIdentify + charcater;
             }

src/main/java/com/yuxue/easypr/core/CharsSegment.java

@@ -124,18 +124,19 @@ public class CharsSegment {
                 CV_RETR_EXTERNAL, // retrieve the external contours
                 CV_CHAIN_APPROX_NONE); // all pixels of each contours
 
-        // Start to iterate to each contour founded
-
         // Remove patch that are no inside limits of aspect ratio and area.
         // 将不符合特定尺寸的图块排除出去
         Vector<Rect> vecRect = new Vector<Rect>();
         for (int i = 0; i < contours.size(); ++i) {
             Rect mr = boundingRect(contours.get(i));
-            /*Mat temp = new Mat(img_threshold, mr);
-            String str = tempPath + "temp_"+i+".jpg";
-            opencv_imgcodecs.imwrite(str, temp);*/
+            Mat contour = new Mat(img_threshold, mr);
+            
+            if (this.isDebug) {
+                String str = tempPath + "debug_char_contour"+i+".jpg";
+                opencv_imgcodecs.imwrite(str, contour);
+            }
             
-            if (verifySizes(new Mat(img_threshold, mr))) {
+            if (verifySizes(contour)) { // 将不符合特定尺寸的图块排除出去
                 vecRect.add(mr);
             }
         }
@@ -149,7 +150,6 @@ public class CharsSegment {
 
         // 获得指示城市的特定Rect,如苏A的"A"
         int specIndex = GetSpecificRect(sortedRect, color);
-        System.err.println(specIndex);
 
         if (this.isDebug) {
             if (specIndex < sortedRect.size()) {

src/test/java/com/yuxue/test/EasyPrTest.java

@@ -1,22 +1,17 @@
 package com.yuxue.test;
 
-import static org.bytedeco.javacpp.opencv_imgproc.CV_CHAIN_APPROX_NONE;
-import static org.bytedeco.javacpp.opencv_imgproc.CV_RETR_EXTERNAL;
-import static org.bytedeco.javacpp.opencv_imgproc.boundingRect;
-import static org.bytedeco.javacpp.opencv_imgproc.findContours;
+import org.bytedeco.javacpp.opencv_imgproc;
 
 import java.io.File;
 import java.util.Vector;
 
 import org.bytedeco.javacpp.opencv_core.Mat;
-import org.bytedeco.javacpp.opencv_core.MatVector;
-import org.bytedeco.javacpp.opencv_core.Rect;
+import org.bytedeco.javacpp.opencv_highgui;
 import org.bytedeco.javacpp.opencv_imgcodecs;
 import org.junit.Test;
 
 import com.yuxue.easypr.core.CharsIdentify;
 import com.yuxue.easypr.core.CharsRecognise;
-import com.yuxue.easypr.core.CharsSegment;
 import com.yuxue.easypr.core.CoreFunc;
 import com.yuxue.easypr.core.PlateDetect;
 import com.yuxue.easypr.core.PlateLocate;
@@ -185,8 +180,8 @@ public class EasyPrTest {
 
     @Test
     public void testGreenPlate() {
-        /*String imgPath = "res/image/test_image/result_0.png";
-        Mat src = opencv_imgcodecs.imread(imgPath);*/
+        String imgPath = "res/image/test_image/debug_resize_2.jpg";
+        Mat src = opencv_imgcodecs.imread(imgPath);
         
         // 获取绿牌的H值范围
         /*MatVector hsvSplit = new MatVector();
@@ -227,12 +222,33 @@ public class EasyPrTest {
         
         // 判断绿色车牌
         /*Mat src_hsv = new Mat();
-        cvtColor(src, src_hsv, CV_BGR2HSV);
+        opencv_imgproc.cvtColor(src, src_hsv, opencv_imgproc.CV_BGR2HSV);
         src_hsv = CoreFunc.colorMatch(src, PlateColor.GREEN, true);
         System.err.println(CoreFunc.plateColorJudge(src, PlateColor.GREEN, true));
         String str = "d:/PlateDetect/src_hsv.png";
         opencv_imgcodecs.imwrite(str, src_hsv);*/
         
+        // 车牌检测对象
+        PlateDetect plateDetect = new PlateDetect();
+        plateDetect.setPDLifemode(true);
+        plateDetect.setDebug(false, ""); // 将过程的图块保存到盘符
+        
+        Vector<Mat> matVector = new Vector<Mat>();
+        
+        System.err.println(plateDetect.plateDetect(src, matVector));
+        System.err.println(matVector.size());
+        
+        for (int i = 0; i < matVector.size(); ++i) { // 遍历车牌图块Mat，进行识别
+            Mat img = matVector.get(i);
+            // 识别的车牌，保存图片文件
+            // 此方法生成的文件，中文名称都是乱码，试了各种编解码均无效，OpenCV自身的编解码问题。
+            opencv_highgui.imshow("123", img);
+            
+            String str = "d:/PlateDetect/temp/result_.png";
+            opencv_imgcodecs.imwrite(str, img);
+            
+        }
+        
     }