- Centrage des caractères beaucoup plus rapide (boucle sur les fichiers dans le dossier faite en C++), mais spécifique windows

- Génération d'un modèles à partir de données simulées centrées avec plus de rotations => 77.5% de succès
- Changement définitif de la manière de tester un modèle sur les captchas Egoshare => beaucoup  plus rapide

Author: remi.paucher

C++ sources/Centrage caractères/main.cpp

@@ -7,6 +7,17 @@
 #include <algorithm>
 #include <sstream>
 
+#include <iostream>
+#include "windows.h"
+#include "stdio.h"
+#include "dos.h" 
+#include <sstream>
+
+#include <cstdlib>
+#include <conio.h>
+
+
+
 #if defined (WIN32)
 #pragma comment(lib,"cv")
 #pragma comment(lib,"cvaux")
@@ -18,24 +29,23 @@
 using namespace std;
 
 
-#define WIDTH 38
-#define HEIGHT 31
-
-
+
+string narrow( wstring& str )
+{
+	ostringstream stm ;
+	const ctype<char>& ctfacet =
+		use_facet< ctype<char> >( stm.getloc() ) ;
+	for( size_t i=0 ; i<str.size() ; ++i )
+		stm << ctfacet.narrow( str[i], 0 ) ;
+	return stm.str() ;
+}
 
 
 
 
-int main(int argc, char *argv[])
+void process_file(string filenameIN, int WIDTH, int HEIGHT)
 {
-	//Noms de fichiers
-	string filenameIN;
-	if (argc < 2)
-		filenameIN = "test.bmp";
-	else
-		filenameIN = argv[1];
-
-
+	//cout << "processing file: " << filenameIN << endl;
 	//Chargement de l'image
 	IplImage *srcImg=0, *res=0;
 	srcImg = cvLoadImage(filenameIN.c_str(),0);
@@ -69,19 +79,111 @@ int main(int argc, char *argv[])
 	{
 		for (int j=0; j<=ymax-ymin; ++j)
 		{
-			cvSet2D(res, offsety+j, offsetx+i, cvGet2D(srcImg, ymin+j, xmin+i));
+			if ((offsety+j>0) && (offsety+j<res->height) && (offsetx+i>0) && (offsetx+i<res->width))
+				cvSet2D(res, offsety+j, offsetx+i, cvGet2D(srcImg, ymin+j, xmin+i));
 		}
 	}
 
+	cvSaveImage(filenameIN.c_str(), res);
+
+}
+
+
+
+
+
+
+
+int main(int argc, char *argv[])
+{
+	//Noms de fichiers
+	string folder;
+
+	//cout << "ARGC: " << argc << endl;
+
+	int WIDTH;
+	int HEIGHT;
+
+	if (argc < 2)
+	{
+		folder = "*";
+		WIDTH = 38;
+		HEIGHT = 31;
+	}
+	else
+	{
+		folder = argv[1];
+		//cout << "filename IN " << folder << endl;
+
+		WIDTH = atoi(argv[2]);
+		//cout << "WIDTH " << WIDTH << endl;
+
+		HEIGHT = atoi(argv[3]);
+		//cout << "HEIGHT " << HEIGHT << endl;
+	}
+
+
+	string expression_bmp = folder + "\\*.bmp";
+	std::vector<std::wstring> SongsLoaded;
+
+	//InitFunction
+	WIN32_FIND_DATA findFileData;
+
+
+
+	char *expr = (char *) expression_bmp.c_str();
+	wchar_t *lpfile = new wchar_t[200];
+	mbstowcs(lpfile, expr, strlen(expr));
+
+	//cout << "sizeof: " << sizeof(expr) << endl;
+	//cout << "EXPRESSION_BMP CHAR*: " << expr << endl;
+	//wcout << "EXPRESSION_BMP: " << lpfile << endl;
+
+	HANDLE hFind = FindFirstFile(lpfile, &findFileData);
+
+	if(hFind  == INVALID_HANDLE_VALUE)
+	{
+		//cout << "Could Not Find Any file in folder!" << endl;
+		//system("pause");
+		return 1;
+	}
+
+	SongsLoaded.push_back(findFileData.cFileName);
+
+	while(FindNextFile(hFind, &findFileData))
+	{
+		SongsLoaded.push_back(findFileData.cFileName);
+		//wcout << findFileData.cFileName << endl;
+	}
+
+	FindClose(hFind);
+
+	for (int i=0; i<SongsLoaded.size(); ++i)
+	{
+		//wcout << "Traitement: " << SongsLoaded[i] << endl;
+
+		string filename = narrow(SongsLoaded[i]);
+		filename = folder + "\\" + filename;
+
+		process_file(filename, WIDTH, HEIGHT);
+
+		//getchar();
+
+	}
+	//cout << endl;
+
+
+
 
-	//cout << "xmin: " << xmin << " xmax: " << xmax << endl;
-	//cout << "ymin: " << ymin << " ymax: " << ymax << endl;
 
-	cvSaveImage(filenameIN.c_str(), res);
 
 	//system("pause");
 
 	return 0;
 
 
 }
+
+
+
+

Egoshare/Models/Stats.txt

@@ -152,4 +152,9 @@ SIMULATION-BASED avec centrage des caract
 simulation_based_C=1000_KERNEL=2.svm	72.8291316527%
 simulation_based_C=1000_KERNEL=1.svm	73.6694677871%
 simulation_based_C=1000_KERNEL=1.svm	73.6694677871%
-simulation_based_C=1000_KERNEL=1.svm	73.6694677871%
+
+Avec plus de rotations:
+-----------------------
+simulation_based_NEW_C=1000_KERNEL=2.svm	76.0504201681%
+simulation_based_NEW_C=1000_KERNEL=1.svm	77.4509803922%
+

Egoshare/Models/simulation_based_C=1000_KERNEL=1.svm

@@ -12,7 +12,7 @@
 if SIMULATION_BASED:
     DEFAULT_SIZE = (30, 30)
     GENERATE_TRAINING_SET = True
-    GENERATE_VALIDATION_SET = True
+    GENERATE_VALIDATION_SET = False
 
     if GENERATE_TRAINING_SET:
         print """
@@ -35,13 +35,12 @@
         SCALE_MIN = 17
         SCALE_MAX = 22
         STEP = 1
-        ALIGN_RANGEY = [0.7]
+        ALIGN_RANGEY = [0.5]
         ALIGN_RANGEX = [0.5]
-        SEUIL_RANGE = [160, 180, 200]
-        ROTATIONS = [2, 4, 6, 9, 13, 17, 20, 22, 25, 27]
-        FONTS = [("Fonts/comic.ttf", (140, 160)),
+        ROTATIONS = [2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30]
+        FONTS = [("Fonts/comic.ttf", (145, 163)),
                  ("Fonts/vera.ttf", (160, 180)),
-                 ("Fonts/califb.ttf", (160, 180))]
+                 ("Fonts/califb.ttf", (171, 191))]
         Generate_Set(DESTINATION_FOLDER,CLEAN_DESTINATION_FOLDER,DISTORTION_W_MIN,DISTORTION_W_MAX,DISTORTION_H_MIN,
                      DISTORTION_H_MAX,SCALE_MIN,SCALE_MAX,STEP, elem_to_gen, FONTS, ALIGN_RANGEX, ALIGN_RANGEY, DEFAULT_SIZE, ROTATIONS)
 

Egoshare_2_GenerateDB.py

@@ -5,8 +5,8 @@
 import time
 
 
-GENERATE_CAPTCHA_BASED_MODELS = True
-GENERATE_SIMULATION_BASED_MODELS = False
+GENERATE_CAPTCHA_BASED_MODELS = False
+GENERATE_SIMULATION_BASED_MODELS = True
 VERBOSE = 0
 MODEL_FOLDER = 'Egoshare/Models'
 
@@ -24,7 +24,7 @@
             #Génération du modèle
             execfile("Train & Test SVM.py")
             #Test du modèle
-            execfile("Egoshare_5_Perf.py")
+            execfile("Egoshare_5_TestPerf.py")
 
 
 if GENERATE_SIMULATION_BASED_MODELS:
@@ -36,8 +36,10 @@
 
     for C in CRANGE:
         for KERNEL in KERNEL_TYPE:
-            MODEL_FILE = "simulation_based_C="+str(C)+"_KERNEL="+str(KERNEL)+".svm"
+            MODEL_FILE = "simulation_based_NEW_C="+str(C)+"_KERNEL="+str(KERNEL)+".svm"
             #Génération du modèle
             execfile("Train & Test SVM.py")
-
+            #Test du modèle
+            execfile("Egoshare_5_TestPerf.py")
+            
 raw_input()

Egoshare_3_TrainTestSVM.py

@@ -10,22 +10,14 @@
 
 for folder, subfolders, files in os.walk(TRAINING_FOLDER):
     loaded = False
-    for file in [file for file in files if 'bmp' in file]:
-        if not loaded:
-            print "folder", folder, "loaded"
-            loaded = True
-        
-        print file
-        
-        if os.name == "nt":
-            filename = os.path.join(os.getcwd(), folder, file)
-            command = '""'+os.path.join(os.getcwd(), 'centrage.exe" "'+filename+'" '+str(WIDTH)+' '+str(HEIGHT)+'"')
-        else:
-            #Some Linux stuff :)
-            pass
-        
-        os.system(command)
-
-
 
+    print folder
 
+    if os.name == "nt":
+        command = 'centrage.exe "%s" %d %d'%(folder, WIDTH, HEIGHT)
+    else:
+        #Some Linux stuff :)
+        pass
+    
+    os.system(command)
+