Image Recognition with Neural Networks

/*##########################################################################
 * 
 * BP1Layer.cs
 * -------------------------------------------------------------------------
 * By
 * Murat FIRAT, June 2007
 * muratti24@gmail.com
 *
 * -------------------------------------------------------------------------
 * Last Update:
 * July,4th 2007
 * 
 * -------------------------------------------------------------------------
 * Description:
 * BP1Layer.cs Implements Single Layer Backpropagation Neural Network
 * 
 * -------------------------------------------------------------------------
 * Notes:
 * If training process takes too long, modify (mostly increase) learning rate 
 * and initial weight.
 * 
 * To modify initial weight, change x(0<x<1) and y(0<y<100) in the following
 * code segment(in InitializeNetwork() function):
 * 
 * PreInputLayer[i].Weights[j] = x +((double)rand.Next(0, y) / 100);
 * 
 * -------------------------------------------------------------------------
 ###########################################################################*/

using System;
using System.Collections.Generic;
using System.Text;
using System.Collections;
using System.IO;
using System.Runtime.Serialization.Formatters.Binary;
using ANNBase;

namespace ANN
{
    class BP1Layer
    {
        private PreInput[] PreInputLayer;
        private Output[] OutputLayer;

        private int PreInputNum;
        private int OutputNum;

        private ArrayList OutputSet;

        private double LearningRate = 0.2;

        private double currentError=99999;
        private bool stopTraining = false;

        public BP1Layer(int preInputNum, int outputNum)
        {
            PreInputNum = preInputNum;
            OutputNum = outputNum;

            PreInputLayer = new PreInput[PreInputNum];
            OutputLayer = new Output[OutputNum];
            OutputSet = new ArrayList();
            InitializeNetwork();
        }

        public void Train(ArrayList TrainingInputs,ArrayList TrainingOutputs,double MaxError)
        {            
            foreach (string s in TrainingOutputs)
            {
                if (!OutputSet.Contains(s))
                {
                    OutputSet.Add(s);
                }
            }
            stopTraining = false;
            do
            {
                for (int i = 0; i < TrainingInputs.Count; i++)
                {
                    ForwardPropagate((double[])TrainingInputs[i], (string)TrainingOutputs[i]);
                    BackPropagate();
                }
                currentError = GetTotalError(TrainingInputs, TrainingOutputs);
            } while (currentError > MaxError && !stopTraining);
            
        }       

        //Apply Input to Network And Find the [Two] Highest Outputs Which are the [Two] Best Matched
        public void Recognize(double[] pattern, ref string MatchedHigh, ref double OutputValueHight, 
                                                ref string MatchedLow,  ref double OutputValueLow)
        {
            int i, j;
            double total = 0.0;
            double max = -1;

            //Apply Input to Network
            for (i = 0; i < PreInputNum; i++)
            {
                PreInputLayer[i].Value = pattern[i];
            }

            //Find the [Two] Highest Outputs
            for (i = 0; i < OutputNum; i++)
            {
                total = 0.0;
                for (j = 0; j < PreInputNum; j++)
                {
                    total += PreInputLayer[j].Value * PreInputLayer[j].Weights[i];
                }
                OutputLayer[i].InputSum = total;
                OutputLayer[i].output = F(total);
                if (OutputLayer[i].output > max)
                {
                    MatchedLow = MatchedHigh;
                    OutputValueLow = max;
                    max = OutputLayer[i].output;
                    MatchedHigh = (string)OutputSet[i];
                    OutputValueHight = max;
                }
            }
        }

        private double F(double x) { return (1 / (1 + Math.Exp(-x))); }

        public double GetTotalError(ArrayList TrainingInputs, ArrayList TrainingOutputs)
        {
            double total = 0.0;
            for (int i = 0; i < TrainingInputs.Count; i++)
            {
                ForwardPropagate((double[])TrainingInputs[i], (string)TrainingOutputs[i]);
                for (int j = 0; j < OutputNum; j++)
                {
                    total += Math.Pow((OutputLayer[j].Target - OutputLayer[j].output), 2) / 2;
                }
            }
            return total;
        }

        public void SaveNetwork(string FileName,int AvHeight, int AvWidth)
        {
            ArrayList AL = new ArrayList();
            AL.Add("1Layer");
            AL.Add(OutputSet);
            AL.Add(PreInputLayer);
            AL.Add(OutputLayer);
            AL.Add(AvHeight);
            AL.Add(AvWidth);
            FileStream FS = new FileStream(FileName, FileMode.OpenOrCreate);
            BinaryFormatter BF = new BinaryFormatter();            
            BF.Serialize(FS,AL);
            FS.Close();
        }

        public void LoadNetwork(string FileName, ref int AvHeight,ref int AvWidth)
        {
            FileStream FS = null;
            try
            {
                FS = new FileStream(FileName, FileMode.Open);
                BinaryFormatter BF = new BinaryFormatter();
                ArrayList AL = (ArrayList)BF.Deserialize(FS);
                string Ver = (string)AL[0];
                if (Ver != "1Layer")
                {
                    throw new Exception("The Loaded Network Does Not Belong To 1 Layer Network But Belongs to " + Ver);
                }
                OutputSet = (ArrayList)AL[1];
                PreInputLayer = (PreInput[])AL[2];
                OutputLayer = (Output[])AL[3];
                AvHeight = (int)AL[4];
                AvWidth = (int)AL[5];
                
                PreInputNum = PreInputLayer.Length;
                OutputNum = OutputLayer.Length;
            }
            finally
            {
                FS.Close();
            }
        }

        //Initialize the Network Weights to Small Values (between 0 and 1) 
        private void InitializeNetwork()
        {
            Random rand = new Random();
            for (int i = 0; i < PreInputNum; i++)
            {
                PreInputLayer[i].Weights = new double[OutputNum];
                for (int j = 0; j < OutputNum; j++)
                {
                    PreInputLayer[i].Weights[j] = 0.01 +((double)rand.Next(0, 2) / 100);
                }
            }
        }

        private void BackPropagate()
        {
            //Update The First Layer's Weights
            for (int j = 0; j < OutputNum; j++)
            {
                for (int i = 0; i < PreInputNum; i++)
                {
                    PreInputLayer[i].Weights[j] += LearningRate * (OutputLayer[j].Error) * PreInputLayer[i].Value;
                }
            }
        }

        private void ForwardPropagate(double[] pattern, string output)
        {
            int i, j;
            double total = 0.0;

            //Apply input to the network
            for (i = 0; i < PreInputNum; i++)
            {
                PreInputLayer[i].Value = pattern[i];
            }

            //Calculate The First(Output) Layer's Inputs, Outputs, Targets and Errors
            for (i = 0; i < OutputNum; i++)
            {
                total = 0.0;
                for (j = 0; j < PreInputNum; j++)
                {
                    total += PreInputLayer[j].Value * PreInputLayer[j].Weights[i];
                }
                OutputLayer[i].InputSum = total;
                OutputLayer[i].output = F(total);
                OutputLayer[i].Target = (((string)OutputSet[i]) == output ? 1.0 : 0.0);
                OutputLayer[i].Error = (OutputLayer[i].Target - OutputLayer[i].output) * (OutputLayer[i].output) * (1 - OutputLayer[i].output);
            }
        }

        public double CurrentError
        {
            get
            {
                return currentError;
            }
        }

        public bool StopTraining
        {
            set
            {
                stopTraining = value;
            }
        }

    }
}