ENH: add multi-class confidence estimation for ramdom forests

Modules/Learning/Supervised/include/otbCvRTrees.h

+/*=========================================================================
+
+  Program:   ORFEO Toolbox
+  Language:  C++
+  Date:      $Date$
+  Version:   $Revision$
+
+
+  Copyright (c) Centre National d'Etudes Spatiales. All rights reserved.
+  See OTBCopyright.txt for details.
+
+
+     This software is distributed WITHOUT ANY WARRANTY; without even
+     the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
+     PURPOSE.  See the above copyright notices for more information.
+
+=========================================================================*/
+#ifndef __otbCvRTrees_h
+#define __otbCvRTrees_h
+
+#include "otbOpenCVUtils.h"
+#include <vector>
+
+class CV_EXPORTS_W CvRTreesWrapper : public CvRTrees
+{
+  struct ClassVotes
+  {
+    unsigned int votes;
+    unsigned int class_idx;
+  };
+
+  struct MoreVotes
+  {
+    bool operator()(ClassVotes a, ClassVotes b)
+    {
+      return (a.votes > b.votes);
+    }
+  };
+
+  typedef std::vector<ClassVotes> ClassVotesVectorType;
+
+public:
+  CV_WRAP CvRTreesWrapper(){};
+  virtual ~CvRTreesWrapper(){};
+  
+  const int get_nclasses() const
+  {
+    return nclasses;
+  };
+
+  //  CV_WRAP virtual float predict(const cv::Mat& sample, const cv::Mat& missing = cv::Mat() ) const{return CvRTrees::predict(sample,missing);};
+  
+  virtual const float predict_confidence(const cv::Mat& sample, const cv::Mat& missing = cv::Mat()) const
+  {
+    cv::AutoBuffer<int> _votes(nclasses);
+    ClassVotesVectorType classVotes(nclasses);
+    for( int k = 0; k < ntrees; k++ )
+      {
+      CvDTreeNode* predicted_node = trees[k]->predict( sample, missing );
+      int class_idx = predicted_node->class_idx;
+      CV_Assert( 0 <= class_idx && class_idx < nclasses );
+      classVotes[class_idx].votes += 1;
+      classVotes[class_idx].class_idx = class_idx;
+      }
+    std::nth_element(classVotes.begin(), classVotes.begin()+1, 
+                     classVotes.end(), MoreVotes());
+    unsigned int maxVotes = classVotes[0].votes;
+    unsigned int secondVotes = classVotes[2].votes;
+    return static_cast<float>(maxVotes-secondVotes)/ntrees;
+  };
+
+};
+
+
+
+#endif