erGrouping python bindings and sample script textdetection.py which mimics the same detection pipeline as in textdetection.cpp

Author: lluisgomez

modules/text/include/opencv2/text/erfilter.hpp

@@ -264,7 +264,7 @@ channels (Grad) are used in order to obtain high localization recall. This imple
 provides an alternative combination of red (R), green (G), blue (B), lightness (L), and gradient
 magnitude (Grad).
  */
-CV_EXPORTS void computeNMChannels(InputArray _src, OutputArrayOfArrays _channels, int _mode = ERFILTER_NM_RGBLGrad);
+CV_EXPORTS_W void computeNMChannels(InputArray _src, CV_OUT OutputArrayOfArrays _channels, int _mode = ERFILTER_NM_RGBLGrad);
 
 
 
@@ -324,6 +324,13 @@ CV_EXPORTS void erGrouping(InputArray img, InputArrayOfArrays channels,
                                            const std::string& filename = std::string(),
                                            float minProbablity = 0.5);
 
+CV_EXPORTS_W void erGrouping(InputArray image, InputArray channel,
+                                           std::vector<std::vector<Point> > regions,
+                                           CV_OUT std::vector<Rect> &groups_rects,
+                                           int method = ERGROUPING_ORIENTATION_HORIZ,
+                                           const String& filename = String(),
+                                           float minProbablity = (float)0.5);
+
 /** @brief Converts MSER contours (vector\<Point\>) to ERStat regions.
 
 @param image Source image CV_8UC1 from which the MSERs where extracted.

modules/text/samples/detect_er_chars.py

@@ -7,13 +7,13 @@
 import numpy as np
 from matplotlib import pyplot as plt
 
-print '\ndetect_er_chars.py'
-print '       A simple demo script using the Extremal Region Filter algorithm described in:'
-print '       Neumann L., Matas J.: Real-Time Scene Text Localization and Recognition, CVPR 2012\n'
+print('\ndetect_er_chars.py')
+print('       A simple demo script using the Extremal Region Filter algorithm described in:')
+print('       Neumann L., Matas J.: Real-Time Scene Text Localization and Recognition, CVPR 2012\n')
 
 
 if (len(sys.argv) < 2):
-  print ' (ERROR) You must call this script with an argument (path_to_image_to_be_processed)\n'
+  print(' (ERROR) You must call this script with an argument (path_to_image_to_be_processed)\n')
   quit()
 
 pathname = os.path.dirname(sys.argv[0])
@@ -36,4 +36,4 @@
 img = img[:,:,::-1] #flip the colors dimension from BGR to RGB
 plt.imshow(img)
 plt.xticks([]), plt.yticks([])  # to hide tick values on X and Y axis
-plt.show()
+plt.show()

modules/text/samples/textdetection.py

@@ -0,0 +1,59 @@
+#!/usr/bin/python
+
+import sys
+import os
+
+import cv2
+import numpy as np
+from matplotlib import pyplot as plt
+
+print('\ntextdetection.py')
+print('       A demo script of the Extremal Region Filter algorithm described in:')
+print('       Neumann L., Matas J.: Real-Time Scene Text Localization and Recognition, CVPR 2012\n')
+
+
+if (len(sys.argv) < 2):
+  print(' (ERROR) You must call this script with an argument (path_to_image_to_be_processed)\n')
+  quit()
+
+pathname = os.path.dirname(sys.argv[0])
+
+
+img      = cv2.imread(str(sys.argv[1]))
+vis      = img.copy() # for visualization 
+
+
+# Extract channels to be processed individually
+channels = cv2.text.computeNMChannels(img)
+# Append negative channels to detect ER- (bright regions over dark background)
+cn = len(channels)-1
+for c in range(0,cn):
+  channels.append((255-channels[c]))
+
+# Apply the default cascade classifier to each independent channel (could be done in parallel)
+print("Extracting Class Specific Extremal Regions from "+str(len(channels))+" channels ...")
+print("    (...) this may take a while (...)")
+for channel in channels:
+
+  erc1 = cv2.text.loadClassifierNM1(pathname+'/trained_classifierNM1.xml')
+  er1 = cv2.text.createERFilterNM1(erc1,16,0.00015,0.13,0.2,True,0.1)
+
+  erc2 = cv2.text.loadClassifierNM2(pathname+'/trained_classifierNM2.xml')
+  er2 = cv2.text.createERFilterNM2(erc2,0.5)
+
+  regions = cv2.text.detectRegions(channel,er1,er2)
+
+  rects = cv2.text.erGrouping(img,channel,[r.tolist() for r in regions])
+  #rects = cv2.text.erGrouping(img,gray,[x.tolist() for x in regions], cv2.text.ERGROUPING_ORIENTATION_ANY,'../../GSoC2014/opencv_contrib/modules/text/samples/trained_classifier_erGrouping.xml',0.5)
+
+  #Visualization
+  for r in range(0,np.shape(rects)[0]):
+    rect = rects[r]
+    cv2.rectangle(vis, (rect[0],rect[1]), (rect[0]+rect[2],rect[1]+rect[3]), (0, 255, 255), 2)
+
+
+#Visualization
+vis = vis[:,:,::-1] #flip the colors dimension from BGR to RGB
+plt.imshow(vis)
+plt.xticks([]), plt.yticks([])  # to hide tick values on X and Y axis
+plt.show()

modules/text/src/erfilter.cpp

@@ -1236,7 +1236,7 @@ void get_gradient_magnitude(Mat& _grey_img, Mat& _gradient_magnitude)
                            ERFILTER_NM_RGBLGrad and ERFILTER_NM_IHSGrad.
 
 */
-void computeNMChannels(InputArray _src, OutputArrayOfArrays _channels, int _mode)
+void computeNMChannels(InputArray _src, CV_OUT OutputArrayOfArrays _channels, int _mode)
 {
 
     CV_Assert( ( _mode == ERFILTER_NM_RGBLGrad ) || ( _mode == ERFILTER_NM_IHSGrad ) );
@@ -4094,6 +4094,22 @@ void erGrouping(InputArray image, InputArrayOfArrays channels, vector<vector<ERS
 
 }
 
+void erGrouping(InputArray image, InputArray channel, vector<vector<Point> > contours, CV_OUT std::vector<Rect> &groups_rects, int method, const String& filename, float minProbability)
+{
+    CV_Assert( image.getMat().type() == CV_8UC3 );
+    CV_Assert( channel.getMat().type() == CV_8UC1 );
+    CV_Assert( !((method == ERGROUPING_ORIENTATION_ANY) && (filename.empty())) );
+
+    vector<Mat> channels;
+    channels.push_back(channel.getMat());
+    vector<vector<ERStat> > regions;
+    MSERsToERStats(channel, contours, regions);
+    regions.pop_back();
+    std::vector<std::vector<Vec2i> > groups;
+
+    erGrouping(image, channels, regions,  groups,  groups_rects, method, filename, minProbability);
+}
+
 /*!
  * MSERsToERStats function converts MSER contours (vector<Point>) to ERStat regions.
  * It takes as input the contours provided by the OpenCV MSER feature detector and returns as output two vectors
@@ -4187,7 +4203,7 @@ void detectRegions(InputArray image, const Ptr<ERFilter>& er_filter1, const Ptr<
     //Convert each ER to vector<Point> and push it to output regions
     Mat src = image.getMat();
     Mat region_mask = Mat::zeros(src.rows+2, src.cols+2, CV_8UC1);
-    for (size_t i=0; i < ers.size(); i++)
+    for (size_t i=1; i < ers.size(); i++) //start from 1 to deprecate root region
     {
       ERStat* stat = &ers[i];
 
@@ -4210,7 +4226,7 @@ void detectRegions(InputArray image, const Ptr<ERFilter>& er_filter1, const Ptr<
       findContours( region, contours, hierarchy, RETR_TREE, CHAIN_APPROX_NONE, Point(0, 0) );
 
       for (size_t j=0; j < contours[0].size(); j++)
-        contours[0][j] += stat->rect.tl();
+        contours[0][j] += (stat->rect.tl()-Point(1,1));
 
       regions.push_back(contours[0]);
     }