Index: modules/core/src/matrix.cpp
===================================================================
--- modules/core/src/matrix.cpp	(revision 6258)
+++ modules/core/src/matrix.cpp	(working copy)
@@ -2196,6 +2196,51 @@
     return d;
 }
 
+#ifdef HAVE_TBB 
+
+class KMeansPPDistanceComputer
+{
+  float *tdist2;
+  const float *data;
+  const float *dist;
+  const int dims;
+  const int step;
+  const int stepci;
+
+public:
+
+  KMeansPPDistanceComputer(float *_tdist2,
+			   const float *_data,
+			   const float *_dist,
+			   int _dims,
+			   int _step,
+			   int _stepci)
+    : tdist2(_tdist2),
+      data(_data),
+      dist(_dist),
+      dims(_dims),
+      step(_step),
+      stepci(_stepci)
+  {
+
+  }
+
+  void operator()(const cv::BlockedRange& range) const
+  {
+
+    const int begin = range.begin();
+    const int end = range.end();
+
+    for (int i=begin; i<end; ++i) {
+      tdist2[i] = std::min(distance(data + step*i, data + stepci, dims), dist[i]);
+    }
+  }
+  
+};
+
+#endif //HAVE_TBB
+
+
 /*
 k-means center initialization using the following algorithm:
 Arthur & Vassilvitskii (2007) k-means++: The Advantages of Careful Seeding
@@ -2227,16 +2272,27 @@
 
         for( j = 0; j < trials; j++ )
         {
-            double p = (double)rng*sum0, s = 0;
+            double p = (double)rng*sum0;
             for( i = 0; i < N-1; i++ )
                 if( (p -= dist[i]) <= 0 )
                     break;
             int ci = i;
+	    double s=0;
+#ifndef HAVE_TBB 
             for( i = 0; i < N; i++ )
             {
                 tdist2[i] = std::min(distance(data + step*i, data + step*ci, dims), dist[i]);
                 s += tdist2[i];
             }
+#else
+	    cv::parallel_for(cv::BlockedRange(0, N),
+			     KMeansPPDistanceComputer(tdist2, data, dist, dims, step, step*ci));
+	    
+            for( i = 0; i < N; i++ )
+            {
+	      s += tdist2[i];
+            }
+#endif //HAVE_TBB
             
             if( s < bestSum )
             {