public void GenerateFeaturesTree(IRaster3DInteger raster, IMaxTreeNode <float> bottom_level_node, float[,] features)
{
//Generate from the leafs downwards
Queue <IMaxTreeNode <float> > queue = new Queue <IMaxTreeNode <float> >();
Stack <IMaxTreeNode <float> > node_stack = new Stack <IMaxTreeNode <float> >();
queue.Enqueue(bottom_level_node);
while (queue.Count != 0)
{
IMaxTreeNode <float> node = queue.Dequeue();
node_stack.Push(node);
foreach (IMaxTreeNode <float> child_node in node.GetNodeChildren())
{
queue.Enqueue(child_node);
}
}
while (node_stack.Count != 0)
{
IMaxTreeNode <float> node = node_stack.Pop();
GenerateFeaturesNode(raster, node, features);
}
}
/** Scale and rotation invariant 3D features
* As in disertation Fred Kawanuka page 79 & Wilkinson Roerding for scale invariant matrix
*/
private void GenerateFeaturesNode(
IRaster3DInteger raster,
IMaxTreeNode <float> node,
float [,] features)
{
int node_index = node.NodeIndex;
// 1 get_size
float size = node.CulmativeRealSize;
// 2 compute geometric moments
int [] elements = node.GetElementIndexArrayNodeReal();
foreach (int element_index in elements)
{
raster.GetElementCoordinatesRBA(element_index, element_coordinates);
features[node_index, IndexSX] += element_coordinates[0];
features[node_index, IndexSY] += element_coordinates[1];
features[node_index, IndexSZ] += element_coordinates[2];
features[node_index, IndexSXX] += element_coordinates[0] * element_coordinates[0];
features[node_index, IndexSYY] += element_coordinates[1] * element_coordinates[1];
features[node_index, IndexSZZ] += element_coordinates[2] * element_coordinates[2];
features[node_index, IndexSXY] += element_coordinates[0] * element_coordinates[1];
features[node_index, IndexSXZ] += element_coordinates[0] * element_coordinates[2];
features[node_index, IndexSYZ] += element_coordinates[1] * element_coordinates[2];
}
foreach (IMaxTreeNode <float> child in node.GetNodeChildren())
{
int child_node_index = child.NodeIndex;
features[node_index, IndexSX] += features[child_node_index, IndexSX];
features[node_index, IndexSY] += features[child_node_index, IndexSY];
features[node_index, IndexSZ] += features[child_node_index, IndexSY];
features[node_index, IndexSXX] += features[child_node_index, IndexSXX];
features[node_index, IndexSYY] += features[child_node_index, IndexSYY];
features[node_index, IndexSZZ] += features[child_node_index, IndexSZZ];
features[node_index, IndexSXY] += features[child_node_index, IndexSXY];
features[node_index, IndexSXZ] += features[child_node_index, IndexSXZ];
features[node_index, IndexSYZ] += features[child_node_index, IndexSYZ];
}
// 2 compute covariance matrix (only the parts we will use note tha)
float x_mean = features[node_index, IndexSX] / size;
float y_mean = features[node_index, IndexSY] / size;
float z_mean = features[node_index, IndexSZ] / size;
covariance_matrix[0] = features[node_index, IndexSXX] - 2 * features[node_index, IndexSX] * x_mean + x_mean * x_mean * size;
covariance_matrix[4] = features[node_index, IndexSYY] - 2 * features[node_index, IndexSY] * y_mean + y_mean * y_mean * size;
covariance_matrix[8] = features[node_index, IndexSZZ] - 2 * features[node_index, IndexSZ] * z_mean + z_mean * z_mean * size;
covariance_matrix[1] = features[node_index, IndexSXY] - features[node_index, IndexSX] * y_mean - features[node_index, IndexSY] * x_mean + x_mean * y_mean
* size;
covariance_matrix[2] = features[node_index, IndexSXZ] - features[node_index, IndexSX] * z_mean - features[node_index, IndexSZ] * x_mean + x_mean * z_mean
* size;
covariance_matrix[5] = features[node_index, IndexSYZ] - features[node_index, IndexSY] * z_mean - features[node_index, IndexSZ] * y_mean + y_mean * z_mean
* size;
;
covariance_matrix[0] = (covariance_matrix[0] + size / 12.0f) / (float)Math.Pow(size, 5.0 / 3.0);
covariance_matrix[4] = (covariance_matrix[4] + size / 12.0f) / (float)Math.Pow(size, 5.0 / 3.0);
covariance_matrix[8] = (covariance_matrix[8] + size / 12.0f) / (float)Math.Pow(size, 5.0 / 3.0);
covariance_matrix[1] = covariance_matrix[1] / (float)Math.Pow(size, 5.0 / 3.0);
covariance_matrix[2] = covariance_matrix[2] / (float)Math.Pow(size, 5.0 / 3.0);
covariance_matrix[5] = covariance_matrix[5] / (float)Math.Pow(size, 5.0 / 3.0);
// CollectionTools.print(covariance_matrix);
// 3 compute eigenvalues
ToolsMathMatrix3Float32.EigenvaluesSymetricRBA(covariance_matrix, eigenvalues);
// compute non-compactness
features[node_index, IndexNonCompactness] = eigenvalues[0] + eigenvalues[1] + eigenvalues[2];
// 4 sort so that e_abs[0] < e_abs[1] < e_abs[2]
eigenvalues[0] = Math.Abs(eigenvalues[0]);
eigenvalues[1] = Math.Abs(eigenvalues[1]);
eigenvalues[2] = Math.Abs(eigenvalues[2]);
Array.Sort(eigenvalues);
// 5 compute d values
float d0 = (float)Math.Sqrt((eigenvalues[0] * 20));
float d1 = (float)Math.Sqrt((eigenvalues[1] * 20));
float d2 = (float)Math.Sqrt((eigenvalues[2] * 20));
// 6 Compute features:
// compute elongation differ
features[node_index, IndexElongation] = eigenvalues[2] / eigenvalues[1];
// compute flasness
features[node_index, IndexFlatness] = eigenvalues[1] / eigenvalues[0];
// compute sparceness
features[node_index, IndexSparceness] = d0 * d1 * d2;
features[node_index, IndexSize] = size;
}