void GlobalFun::computeEigenWithTheta(CMesh* _samples, double radius)
{
vector<vector<int> > neighborMap;
typedef vector<CVertex>::iterator VertexIterator;
VertexIterator begin = _samples->vert.begin();
VertexIterator end = _samples->vert.end();
neighborMap.assign(end - begin, vector<int>());
int curr_index = 0;
for (VertexIterator iter=begin; iter!=end; iter++, curr_index++)
{
if(iter->neighbors.size() <= 3)
{
iter->eigen_confidence = 0.5;
continue;
}
for(int j = 0; j < iter->neighbors.size(); j++)
{
neighborMap[curr_index].push_back(iter->neighbors[j]);
}
}
double radius2 = radius*radius;
double iradius16 = -1/radius2;
int currIndex = 0;
for (VertexIterator iter=begin; iter!=end; iter++, currIndex++)
{
if(iter->neighbors.size() <= 3)
{
iter->eigen_confidence = 0.5;
continue;
}
Matrix33d covariance_matrix;
Point3f diff;
covariance_matrix.SetZero();
int neighborIndex = -1;
int neighbor_size = iter->neighbors.size();
for (unsigned int n=0; n<neighbor_size; n++)
{
neighborIndex = neighborMap[currIndex][n];
if(neighborIndex < 0)
break;
VertexIterator neighborIter = begin + neighborIndex;
diff = iter->P() - neighborIter->P();
Point3f vm = iter->N();
Point3f tm = neighborIter->N();
double dist2 = diff.SquaredNorm();
double theta = exp(dist2*iradius16);
for (int i=0; i<3; i++)
for (int j=0; j<3; j++)
covariance_matrix[i][j] += diff[i]*diff[j] * theta;
}
Point3f eigenvalues;
Matrix33d eigenvectors;
int required_rotations;
vcg::Jacobi< Matrix33d, Point3f >(covariance_matrix, eigenvalues, eigenvectors, required_rotations);
vcg::SortEigenvaluesAndEigenvectors< Matrix33d, Point3f >(eigenvalues, eigenvectors);
double sum_eigen_value = (eigenvalues[0] + eigenvalues[1] + eigenvalues[2]);
iter->eigen_confidence = eigenvalues[0] / sum_eigen_value;
for (int d=0; d<3; d++)
iter->eigen_vector0[d] = eigenvectors[d][0];
for (int d=0; d<3; d++)
iter->eigen_vector1[d] = eigenvectors[d][1];
for (int d=0; d<3; d++)
iter->N()[d] = eigenvectors[d][2];
iter->eigen_vector0.Normalize();
iter->eigen_vector1.Normalize();
iter->N().Normalize();
}
}
void GlobalFun::computeEigenIgnoreBranchedPoints(CMesh* _samples)
{
vector<vector<int> > neighborMap;
typedef vector<CVertex>::iterator VertexIterator;
VertexIterator begin = _samples->vert.begin();
VertexIterator end = _samples->vert.end();
neighborMap.assign(end - begin, vector<int>());
int curr_index = 0;
for (VertexIterator iter=begin; iter!=end; ++iter, curr_index++)
{
if(iter->neighbors.size() <= 3)
{
iter->eigen_confidence = 0.5;
continue;
}
//neighborMap[curr_index].push_back(curr_index);
for(int j = 0; j < iter->neighbors.size(); j++)
{
CVertex& t = _samples->vert[iter->neighbors[j]];
if (t.is_skel_branch || t.is_skel_ignore)
{
continue;
}
neighborMap[curr_index].push_back(iter->neighbors[j]);
}
}
int currIndex = 0;
for (VertexIterator iter=begin; iter!=end; iter++, currIndex++)
{
int neighbor_size = neighborMap[currIndex].size();
if (neighbor_size < 3)
{
iter->eigen_confidence = 0.95;
iter->eigen_vector0 = Point3f(0, 0, 0);
continue;
}
Matrix33d covariance_matrix;
Point3f diff;
covariance_matrix.SetZero();
int neighborIndex = -1;
for (unsigned int n=0; n<neighbor_size; n++)
{
neighborIndex = neighborMap[currIndex][n];
if(neighborIndex < 0)
break;
VertexIterator neighborIter = begin + neighborIndex;
diff = iter->P() - neighborIter->P();
for (int i=0; i<3; i++)
for (int j=0; j<3; j++)
covariance_matrix[i][j] += diff[i]*diff[j];
}
Point3f eigenvalues;
Matrix33d eigenvectors;
int required_rotations;
vcg::Jacobi< Matrix33d, Point3f >(covariance_matrix, eigenvalues, eigenvectors, required_rotations);
vcg::SortEigenvaluesAndEigenvectors< Matrix33d, Point3f >(eigenvalues, eigenvectors);
double sum_eigen_value = (eigenvalues[0] + eigenvalues[1] + eigenvalues[2]);
iter->eigen_confidence = eigenvalues[0] / sum_eigen_value;
for (int d=0; d<3; d++)
iter->eigen_vector0[d] = eigenvectors[d][0];
for (int d=0; d<3; d++)
iter->eigen_vector1[d] = eigenvectors[d][1];
for (int d=0; d<3; d++)
iter->N()[d] = eigenvectors[d][2];
iter->eigen_vector0.Normalize();
iter->eigen_vector1.Normalize();
iter->N().Normalize();
}
}