/// <inheridoc/>
public ISignerModel Train(List <Signature> signatures)
{
var genuines = signatures.Where(s => s.Origin == Origin.Genuine)
.Select(s => new { s.ID, Features = s.GetAggregateFeature(Features).ToArray() }).ToList();
var distanceMatrix = new DistanceMatrix <string, string, double>();
foreach (var i in genuines)
{
foreach (var j in genuines)
{
if (distanceMatrix.ContainsKey(j.ID, i.ID))
{
distanceMatrix[i.ID, j.ID] = distanceMatrix[j.ID, i.ID];
}
else if (i == j)
{
distanceMatrix[i.ID, j.ID] = 0;
}
else
{
distanceMatrix[i.ID, j.ID] = DtwPy.Dtw(i.Features, j.Features, DistanceFunction);
}
}
}
var distances = distanceMatrix.GetValues().Where(v => v != 0);
var mean = distances.Average();
var stdev = Math.Sqrt(distances.Select(d => (d - mean) * (d - mean)).Sum() / distances.Count());
double med;
var orderedDistances = new List <double>(distances).OrderBy(d => d);
if (orderedDistances.Count() % 2 == 0)
{
int i = orderedDistances.Count() / 2;
med = (orderedDistances.ElementAt(i - 1) + orderedDistances.ElementAt(i)) / 2.0;
}
else
{
int i = orderedDistances.Count() / 2;
med = orderedDistances.ElementAt(i);
}
return(new DtwSignerModel
{
GenuineSignatures = genuines.Select(g => new KeyValuePair <string, double[][]>(g.ID, g.Features)).ToList(),
DistanceMatrix = distanceMatrix,
Threshold = mean + MultiplicationFactor * stdev
//Threshold = med + (distances.Max() - med) * 1 / 2
});
}
/// <inheridoc/>
public double Test(ISignerModel model, Signature signature)
{
var dtwModel = (DtwSignerModel)model;
var testSignature = signature.GetAggregateFeature(Features).ToArray();
var distances = new double[dtwModel.GenuineSignatures.Count];
for (int i = 0; i < dtwModel.GenuineSignatures.Count; i++)
{
distances[i] = DtwPy.Dtw(dtwModel.GenuineSignatures[i].Value, testSignature, DistanceFunction);
dtwModel.DistanceMatrix[signature.ID, dtwModel.GenuineSignatures[i].Key] = distances[i];
}
// returns value between 0 and 1, how confident is the decision about genuineness
// 1 -> conident genuine, 0 -> not confident genuine = confident forged
return(Math.Max(1 - (distances.Average() / dtwModel.Threshold) / 2, 0));
}
public double[,] Calculate()
{
var distances = new double[InputSignatures.Count, InputSignatures.Count];
for (int i = 0; i < InputSignatures.Count; i++)
{
for (int j = 0; j < InputSignatures.Count; j++)
{
distances[i, j] =
DtwPy.Dtw <double[]>(
InputSignatures[i].GetAggregateFeature(InputFeatures),
InputSignatures[j].GetAggregateFeature(InputFeatures),
InputFunc);
}
}
return(distances);
}
private Tuple <int, Stroke, double, int> CalculateIdx(Stroke stroke, List <Vertex> trajectory, int window, int jump = 1)
{
int resIdx = int.MaxValue;
double resVal = Double.MaxValue;
for (int i = 0; i < trajectory.Count - window; i += jump)
{
double val = DtwPy.Dtw <double[]>(stroke.GetDtwPairingFeature(InputScale),
trajectory.GetRange(i, window).GetDtwPairingFeature(InputScale),
DtwPairingDistance);
if (val < resVal)
{
resVal = val;
resIdx = i;
}
}
return(new Tuple <int, Stroke, double, int>(resIdx, stroke, resVal, window));
}
public List <Tuple <string, string, double> > Calculate()
{
var distances = new double[OfflineSignatures.Count];
Parallel.For(0, OfflineSignatures.Count,
idx =>
{
var offlineSig = OfflineSignatures[idx];
var offlineTrajectory = offlineSig.GetFeature <List <Vertex> >(InputOfflineTrajectory);
var onlineSig = OnlineSignatures.Find(sig => sig.ID == offlineSig.ID && sig.Signer.ID == offlineSig.Signer.ID);
var onlineTrajectory = onlineSig.GetFeature <List <Vertex> >(InputOnlineTrajectory);
distances[idx] = DtwPy.Dtw(offlineTrajectory.GetOriginalXYs(), onlineTrajectory.GetOriginalXYs(), PenDistance);
}
);
var res = new List <Tuple <string, string, double> >();
for (int i = 0; i < OfflineSignatures.Count; i++)
{
res.Add(new Tuple <string, string, double>(OfflineSignatures[i].Signer.ID, OfflineSignatures[i].ID, distances[i]));
}
return(res);
}
public static double GetCost(Signature sig1, Signature sig2, DtwType dtwType, List <FeatureDescriptor> features)
{
switch (dtwType)
{
case DtwType.DtwPy:
return(DtwPy.Dtw(sig1.GetAggregateFeature(features), sig2.GetAggregateFeature(features), DtwPy.EuclideanDistance));
case DtwType.NDtw:
//hatalmas hack, csak egy-egy featurere működik, kombinációra nem
return(new NDtw.Dtw(sig1.GetFeature <List <double> >(features[0]).ToArray(), sig2.GetFeature <List <double> >(features[0]).ToArray(),
DistanceMeasure.Manhattan).GetCost());
case DtwType.FrameworkDtw:
return(new SigStat.Common.Algorithms.Dtw(Accord.Math.Distance.Manhattan)
.Compute(sig1.GetAggregateFeature(features).ToArray(), sig2.GetAggregateFeature(features).ToArray()));
case DtwType.MyDtw:
return(new Common.Dtw(sig1, sig2, features).CalculateDtwScore());
default:
throw new ArgumentException(nameof(dtwType) + typeof(DtwType) + "not exists");
}
}
