private Line2 ApproximateFromTop()
{
//guess axis with top face(s)
Line2 main_axis = null;
face_center = new List <Vector2>();
List <Vector2> normal = new List <Vector2>();
List <double> minoraxislength = new List <double>();
foreach (Image <Gray, byte> fimg in face_img)
{
Vector2 mean = new Vector2();
List <Vector2> points = IExtension.GetMaskPoints(fimg);
foreach (Vector2 p in points)
{
mean += p;
}
mean /= points.Count;
face_center.Add(mean);
float W = points.Count;
float[,] C = new float[2, 2];
foreach (Vector2 point in points)
{
C[0, 0] += (point.x - mean.x) * (point.x - mean.x);
C[0, 1] += (point.x - mean.x) * (point.y - mean.y);
C[1, 0] += (point.y - mean.y) * (point.x - mean.x);
C[1, 1] += (point.y - mean.y) * (point.y - mean.y);
}
C[0, 0] /= W;
C[0, 1] /= W;
C[1, 0] /= W;
C[1, 1] /= W;
Matrix2d CM = new Matrix2d(C);
NumericalRecipes.SVD svd = new NumericalRecipes.SVD(C);
//svd.w - eigenvalue, start from 1
//svd.u - eigenvector, start from 1
int max = 1, min = 2;
if (svd.w[max] < svd.w[min])
{
int temp = max;
max = min;
min = temp;
}
float major = 2 * Mathf.Sqrt(svd.w[max]);
Vector2 majoraxis = new Vector2(svd.u[1, max], svd.u[2, max]);
majoraxis.Normalize();
float minor = 2 * Mathf.Sqrt(svd.w[min]);
Vector2 minoraxis = new Vector2(svd.u[1, min], svd.u[2, min]);
minoraxis.Normalize();
Vector2 majorendp = mean + majoraxis * major;
Vector2 majorstartp = mean - majoraxis * major;
Vector2 minorendp = mean + minoraxis * minor;
Vector2 minorstartp = mean - minoraxis * minor;
//minoraxislength.Add(Vector2.Distance(minorendp, minorstartp));
minoraxislength.Add((double)minor * 2);
normal.Add(Utility.PerpendicularRight(majoraxis));
//normal.Add(majoraxis);
}
if (face_center.Count > 1)
{
//Vector2 mean_normal = normal.First();
Line2 best_top_normal_line = new Line2(face_center.First(), normal.First(), true);
double maxdis = 0;
int farthesttopidx = 0;
for (int i = 1; i < face_center.Count; i++)
{
double tofirstcenterdis = Vector2.Distance(face_center[0], face_center[i]);
if (tofirstcenterdis > maxdis)
{
maxdis = tofirstcenterdis;
farthesttopidx = i;
}
double dis = best_top_normal_line.DistanceToLine(face_center[i]);
if (dis > 10)
{
main_axis = null;
}
}
main_axis = new Line2(face_center.First(), face_center[farthesttopidx]);
}
else
{
Line2 top_normal_line = new Line2(face_center.First(), normal.First(), true);
Vector2 online_point = top_normal_line.GetPointwithT((float)minoraxislength[0]);
if (online_point.x >= 0 && online_point.y >= 0 && online_point.x < body_img.Width && online_point.y < body_img.Height)
{
if (!this.body_img[(int)online_point.y, (int)online_point.x].Equals(new Gray(255)))
{
top_normal_line.Flip();
}
}
else
{
top_normal_line.Flip();
}
main_axis = top_normal_line;
}
return(main_axis);
}
public double[] SVDSingularMatW()
{
SVD svd = new SVD(e, row_size, row_size);
if (svd.State == false) throw new ArithmeticException();
return svd.w;
}
public Matrix4d Inverse()
{
SVD svd = new SVD(e, row_size, row_size);
if (svd.State == false) throw new ArithmeticException();
return new Matrix4d(svd.Inverse);
}
public double[] SVDSingularMat()
{
SVD svd = new SVD(e, 3, 3);
return svd.w;
}
public Matrix3d OrthogonalFactorSVD()
{
SVD svd = new SVD(e, 3, 3);
lastSVDIsFullRank = svd.FullRank;
return new Matrix3d(svd.OrthogonalFactor);
}
public Matrix3d InverseSVD()
{
SVD svd = new SVD(e, 3, 3);
Matrix3d inv = new Matrix3d(svd.Inverse);
lastSVDIsFullRank = svd.FullRank;
return inv;
}