// *************************************************************************************************
// @fn Recognizer
// @brief
// @return
// *************************************************************************************************
public static double Recognizer(SVSystem Template, SVSystem Tester)
{
double Score = 0;
double[] D = new double[FeatureNum];
//-- Feature std
double X_std_T = CalculateFeatureStd(Template.Feature_X);
double Y_std_T = CalculateFeatureStd(Template.Feature_Y);
double P_std_T = CalculateFeatureStd(Template.Feature_P);
double Vx_std_T = CalculateFeatureStd(Template.Feature_Vx);
double Vy_std_T = CalculateFeatureStd(Template.Feature_Vy);
double Ax_std_T = CalculateFeatureStd(Template.Feature_Ax);
double Ay_std_T = CalculateFeatureStd(Template.Feature_Ay);
double A_std_T = CalculateFeatureStd(Template.Feature_A);
double R_std_T = CalculateFeatureStd(Template.Feature_R);
//-- 辨識器
D[0] = ELCSS(Template.Feature_X, Tester.Feature_X, Math.Abs(X_std_T) * 0.65, 5);
D[1] = ELCSS(Template.Feature_Y, Tester.Feature_Y, Math.Abs(Y_std_T) * 0.5, 5);
D[2] = ELCSS(Template.Feature_P, Tester.Feature_P, Math.Abs(P_std_T) * 0.9, 5);
D[3] = ELCSS(Template.Feature_Vx, Tester.Feature_Vx, Math.Abs(Vx_std_T) * 0.55, 5);
D[4] = ELCSS(Template.Feature_Vy, Tester.Feature_Vy, Math.Abs(Vy_std_T) * 0.65, 5); // 0.4
D[5] = ELCSS(Template.Feature_Ax, Tester.Feature_Ax, Math.Abs(Ax_std_T) * 0.95, 5); // 0.9
D[6] = ELCSS(Template.Feature_Ay, Tester.Feature_Ay, Math.Abs(Ay_std_T) * 0.6, 5); // 0.5
D[7] = ELCSS(Template.Feature_A, Tester.Feature_A, Math.Abs(A_std_T) * 0.65, 5);
Similarity = D;
Score = D.Sum() * FeatureNum * 100 / Math.Pow(SVSystem.FeatureNum, 2);;
return(Score);
}
// *************************************************************************************************
// @fn SignatureTemplateSelection
// @brief
// @return
// *************************************************************************************************
public static double[] SignatureTemplateSelection(SVSystem[] Database, ref SVSystem TemplateData, ref List <int> BadSig)
{
//-- 安全分數設定 - 隨機仿冒者分數落在60-70分居多
//int SerurityScore_Norm = 80;
int SerurityScore_Min = 70;
//-- Each person entroll fixed signatures to choose template
int Entrollment = Database.Length;
//-- BestTemplate
double[] BestTemplate = new double[10 + 2 * Entrollment];
//-- Training data
int TrainData_startPos = 0;
//-- Record all signature inner score
double[,] Score_All = new double[Entrollment, Entrollment - 1];
//-- Final score to select which best
double[] TemplateScore = new double[Entrollment];
double[] TemplateScoreMean = new double[Entrollment];
double[] TemplateScoreStd = new double[Entrollment];
for (int i = TrainData_startPos; i < (TrainData_startPos + Entrollment); i++)
{
int cnt = 0;
for (int j = TrainData_startPos; j < (TrainData_startPos + Entrollment); j++)
{
if (i == j)
{
continue;
}
double Score_t = 0;
double[] D = new double[FeatureNum];
Score_t = Recognizer(Database[i], Database[j]);
Score_All[i, cnt++] = Score_t;
}
}
for (int i = TrainData_startPos; i < (TrainData_startPos + Entrollment); i++)
{
//-- 取平均值
double mean = 0;
for (int j = 0; j < (Entrollment - 1); j++)
{
mean += Score_All[i, j];
}
mean = mean / (Entrollment - 1);
//-- 計算平方和
double sum = 0;
for (int k = 0; k < (Entrollment - 1); k++)
{
sum += Math.Pow(Score_All[i, k] - mean, 2);
}
//-- 計算標準差
double std = 0;
std = Math.Sqrt(sum / (Entrollment - 2));
TemplateScore[i] = mean - std;
TemplateScoreMean[i] = mean;
TemplateScoreStd[i] = std;
BestTemplate[2 * i + 10] = mean;
BestTemplate[2 * i + 1 + 10] = std;
}
//-- 選擇最大樣板分數，與最大標準差
double ScoreMax = 0, StdMax = 0;
ScoreMax = TemplateScore.Max();
StdMax = TemplateScoreStd.Max();
for (int i = 0; i < Entrollment; i++)
{
if (TemplateScore[i] == ScoreMax)
{
BestTemplate[0] = i + 1;
BestTemplate[1] = ScoreMax;
BestTemplate[2] = TemplateScoreMean[i];
BestTemplate[3] = TemplateScoreStd[i];
BestTemplate[4] = TemplateScoreMean[i] + 2 * TemplateScoreStd[i];
BestTemplate[5] = TemplateScoreMean[i] - 2 * TemplateScoreStd[i];
//BestTemplate[4] = TemplateScoreMean[i] + StdMax;
//BestTemplate[5] = TemplateScoreMean[i] - StdMax;
BestTemplate[6] = Database[i].CostTime;
break;
}
}
//-- 將相似度分數轉換至0-100分範圍
for (int i = 1; i < BestTemplate.Length; i++)
{
//-- 書寫時間跳過
if (i == 6)
{
continue;
}
BestTemplate[i] = BestTemplate[i] * 100 / Math.Pow(SVSystem.FeatureNum, 2);
//-- 小於 最低安全分數者 分當作絕對仿冒者 or 不好的簽名
if (BestTemplate[i] < SerurityScore_Min && i >= 10 && (i % 2 == 0))
{
BestTemplate[i] = 0;
BestTemplate[i + 1] = 0;
}
//else if (BestTemplate[i] > SerurityScore_Min && BestTemplate[i] < SerurityScore_Norm && i >= 10 && (i % 2 == 0))
//{
// BestTemplate[i] = ((BestTemplate[i] - SerurityScore_Min) / (SerurityScore_Norm - SerurityScore_Min)) * (SerurityScore_Norm - 50) + 50;
//}
}
double Upper = 0;
double lower = 0;
double Mean = 0;
double Std = 0;
Mean = BestTemplate[2];
Std = BestTemplate[3];
Upper = Mean + Std;
lower = Mean - Std;
//-- 樣板標準判斷
if (Std >= 3.5)
{
bool BigBadSig_flag = false;
for (int i = 0; i < Entrollment; i++)
{
if (i == (BestTemplate[0] - 1))
{
continue;
}
if (Math.Abs(BestTemplate[2 * i + 10] - Mean) > 5)
{
BadSig.Add(i + 1);
BigBadSig_flag = true;
}
}
if (BigBadSig_flag == false)
{
for (int i = 0; i < Entrollment; i++)
{
if (i == (BestTemplate[0] - 1))
{
continue;
}
//-- 設定標準差小於閥值建議重新簽名
if (Math.Abs(BestTemplate[2 * i + 1 + 10]) > 5)
{
BadSig.Add(i + 1);
}
//-- 設定小於最佳樣版的最低分數者且該標準差大於閥值，建議重新簽名
//if ((BestTemplate[2 * i + 10] - BestTemplate[2 * i + 1 + 10]) < lower && BestTemplate[2 * i + 1 + 10] >= 3)
//{
// BadSig.Add(i + 1);
//}
}
}
}
//-- 取得最新樣板並回傳
TemplateData = Database[(int)BestTemplate[0] - 1];
return(BestTemplate);
}
