Algorithm A1: Synthesis and training of the correlation neuron

Initiate C-, C+, AUCMAX, b //Select an array Ψ of random pairs of features (not involved in any neuron) //with a cross-correlation level of more than C- but less than C+ for ι from 1 to η do Ψ[ι] = GetRandomUniqueFeaturesPair(C-; C+) end p = 0.9 //Calculation of an array W of neurons’ weight coefficients using Formula (18) //Calculation of meta-features is performed according to Formula (12) for ι from 1 to η do W[ι] = GetSynapsesWeights(Ψ[ι], p) end //Calculation of arrays YG and YI of neuron responses //to «Genuine» and «Impostors» training images according to Formula (17) for k from 1 to KG do YG[k] = GetNeuronResponse(āG,k, Ψ, W, p) end for k from 1 to KI do YI[k] = GetNeuronResponse(āI,k, Ψ, W, p) end ξG = GetMathematicalExpectation(YG) ςG = GetStandardDeviation(YG, ξG) ξI = GetMathematicalExpectation(YI) ςI = GetStandardDeviation(YI, ξI) yImin = ξI − 4ςI yImax = ξI + 4ςI ΔyG = 4ςG yGmin = ξG − ΔyG yGmax = ξG + ΔyG FGmax = F(yGmax)//Formula (21) FGmin = F(yGmin)//Formula (21) ΔFG = FGmax − FGmin if AUCMAX > AUC(ՓG(y), ՓI(y)) then//Formula (22)  do  if ΔFG < 0.4 then   do   while ΔFG < 0.1    do    ΔyG = 1.05ΔyG     yGmin = ξG − ΔyG    yGmax = ξG + ΔyG     FGmax = F(yGmax)//Formula (21)     FGmin = F(yGmin)//Formula (21)    ΔFG = FGmax-FGmin    end   if FGmin < 0.1 then    if 1 − FGmax > 0.6 then     do      ɸG = 3      Tleft = yGmax      ΔT = (yImax − yGmax)/4      Tmiddle = Tleft + ΔT      Tright = Tmiddle + ΔT     H = GetHashTransformationNumber(ɸG, b)     return {H, Ψ, W, Tleft, Tmiddle, Tright}     end    else     return NULL//Neuron is not trained and must be removed   else    if FGmin > 0.4 then     if 1 − FGmax < 0.1 then      if 1 − FGmin < 0.4 then       do        ɸG = 0        ΔT = (yGmin − yImin)/4        Tleft = yGmin − 2ΔT        Tmiddle = yGmin − ΔT        Tright = yGmin        H = GetHashTransformationNumber(ɸG, b)        return {H, Ψ, W, Tleft, Tmiddle, Tright}        end     else        return NULL//Neuron is not trained and must be removed    else     do      ɸG = 1      ΔT = (yGmin − yImin)/3      Tleft = yGmin − ΔT      Tmiddle = yGmin      Tright = yGmax     H = GetHashTransformationNumber(ɸG, b)     return {H, Ψ, W, Tleft, Tmiddle, Tright}     end     else        do         ɸG = 2         ΔT = (yImax-yGmax)/3         Tleft = yGmin         Tmiddle = yGmax         Tright = Tmiddle + ΔT        H = GetHashTransformationNumber(ɸG, b)        return {H, Ψ, W, Tleft, Tmiddle, Tright}       end   end  else   return NULL//Neuron is not trained and must be removed  end else  return NULL//Neuron is not trained and must be removed