Initial commit for Conjugate Gradient method

Implementation is based on
http://www.cs.cmu.edu/~quake-papers/painless-conjugate-gradient.pdf

So far we only have basic interface and empty test. But it compiles at
least.

modules/optim/doc/optim.rst

@@ -10,3 +10,4 @@ optim. Generic numerical optimization
     linear_programming
     downhill_simplex_method
     primal_dual_algorithm
+    conjugate_gradient

modules/optim/include/opencv2/optim.hpp

@@ -54,7 +54,6 @@ public:
     {
     public:
        virtual ~Function() {}
-       //! ndim - dimensionality
        virtual double calc(const double* x) const = 0;
     };
 
@@ -86,6 +85,23 @@ CV_EXPORTS_W Ptr<DownhillSolver> createDownhillSolver(const Ptr<Solver::Function
         InputArray initStep=Mat_<double>(1,1,0.0),
         TermCriteria termcrit=TermCriteria(TermCriteria::MAX_ITER+TermCriteria::EPS,5000,0.000001));
 
+//! conjugate gradient method
+class CV_EXPORTS ConjGradSolver : public Solver
+{
+public:
+    class CV_EXPORTS Function : public Solver::Function
+    {
+    public:
+       //! gradient is like the first derivative for multivar function
+       virtual void getGradient(const double* x,double* buf)const=0;
+       //! Jacobian is like the second derivative
+       virtual void getJacobian(const double* x)const=0;
+    };
+};
+
+CV_EXPORTS_W Ptr<ConjGradSolver> createConjGradSolver(const Ptr<Solver::Function>& f=Ptr<ConjGradSolver::Function>(),
+        TermCriteria termcrit=TermCriteria(TermCriteria::MAX_ITER+TermCriteria::EPS,5000,0.000001));
+
 //!the return codes for solveLP() function
 enum
 {