int OtsuThresholding<T>::computeThreshold(const Vector<unsigned int>& histogram) const
{
const Vector<double> criterion = computeCriterion( histogram );
return criterion.maxPos();
#if 0
double s0, ss0, m0, v0, s1, ss1, m1, v1, vmin, vgng;
// initialisation des statistiques de la classe 0
m0 = s0 = ss0 = v0 = 0;
// initialisation des statistiques de la classe 1
s1 = ss1 = 0;
for (g = 0; g < histogram.getSize(); ++g)
{
n1 += histogram[g];
s1 += g * histogram[g];
ss1 += g * g * histogram[g];
}
m1 = s1 / n1;
v1 = ss1 / n1 - m1 * m1;
vmin = v1;
gmin = 0;
for (g = 0; g < histogram.getSize(); ++g)
{
n0 += histogram[g];
n1 -= histogram[g];
vgng = g * histogram[g];
s0 += vgng;
s1 -= vgng;
ss0 += g * vgng;
ss1 -= g * vgng;
if ( n0 && n1 )
{
m0 = s0 / n0;
m1 = s1 / n1;
v0 = ss0 / n0 - m0 * m0;
v1 = ss1 / n1 - m1 * m1;
if ( v0 + v1 < vmin )
{
vmin = v0 + v1;
gmin = g + 1;
}
}
}
return gmin;
#endif
}
double arlCore::ICP::optimiseLS( void )
{
unsigned int i;
m_startError = m_endError;
arlCore::OptimiseICP_LS optimizer( *this, vnl_least_squares_function::use_gradient );
//optimizer.setObserver(true);
vnl_levenberg_marquardt lm(optimizer);
const arlCore::vnl_rigid_vector V( m_solution );
vnl_vector<double> init(6);
for( i=0 ; i<6 ; ++i )
init(i) = V[i];
lm.minimize_using_gradient(init);
//m_startError = lm.get_start_error();
//m_endError = lm.get_end_error();
m_nbIterations = lm.get_num_iterations();
//m_nbIterations = optimizer.getNbIterations();
m_solution.copy(arlCore::vnl_rigid_vector(init));
m_endError = computeCriterion( m_solution );
m_solution.setRMS(m_endError);
return m_endError;
}
double arlCore::ICP::computeCriterion( const arlCore::vnl_rigid_matrix &M )
{
vnl_vector< double > fx;
return computeCriterion( M, fx );
}
