public LDA(double[][] trainingSet, List<string> labels,int numOfComponents)
{
int n = trainingSet.Length; // sample size
HashSet<String> tempSet = new HashSet<String>(labels);
int c = tempSet.Count; // class size
// process in PCA
PCA pca = new PCA(trainingSet, labels, n - c);
// classify
double[][] meanTotal = new double[n - c][];
for (int i = 0; i < n - c; i++)
{
meanTotal[i] = new double[1];
}
Dictionary<String, List<double[]>> dict = new Dictionary<String, List<double[]>>();
List<projectedTrainingMatrix> pcaTrain = pca.getProjectedTrainingSet();
for (int i = 0; i < pcaTrain.Count; i++) {
String key = pcaTrain[i].label;
meanTotal= meanTotal.Add(pcaTrain[i].matrix);
if (!dict.ContainsKey(key)) {
List<double[]> temp = new List<double[]>();
temp.Add(pcaTrain[i].matrix.Transpose()[0]);
dict.Add(key, temp);
} else {
List<double[]> temp = dict[key];
temp.Add(pcaTrain[i].matrix.Transpose()[0]);
dict[key]= temp;
}
}
meanTotal.ToMatrix().Multiply((double) 1 / n);
// calculate Sw, Sb
double[][] Sw = new double[n - c][];
double[][] Sb = new double[n - c][];
for (int i = 0; i < n - c; i++)
{
Sw[i] = new double[n-c];
Sb[i] = new double[n-c];
}
List<String> labelSet = dict.Keys.ToList();
foreach(string label in labelSet)
{
List<double[]> tempMatrix = dict[label];
double[][] matrixWithinThatClass = tempMatrix.ToArray();
double[] meanOfCurrentClass = Accord.Statistics.Tools.Mean(matrixWithinThatClass);
for (int i = 0; i < matrixWithinThatClass.Length; i++) {
double[][] temp1 = matrixWithinThatClass[i].ToArray().Subtract(meanOfCurrentClass.ToArray());
temp1 = temp1.Multiply(temp1.Transpose());
Sw =Sw.Add(temp1);
}
double[][] temp = meanOfCurrentClass.ToArray().Subtract(meanTotal);
temp = temp.Multiply(temp.Transpose()).ToMatrix().Multiply((double)matrixWithinThatClass.Length).ToArray();
Sb=Sb.Add(temp);
}
// calculate the eigenvalues and vectors of Sw^-1 * Sb
double [][] targetForEigen = Sw.Inverse().Multiply(Sb);
var feature = new EigenvalueDecomposition(targetForEigen.ToMatrix());
double[] d = feature.RealEigenvalues;
int[] indexes;
d.StableSort(out indexes);
indexes = indexes.Reverse().ToArray();
indexes = indexes.Submatrix(0, c - 1);
//int[] indexes = getIndexesOfKEigenvalues(d, c - 1);
double[][] eigenVectors = feature.Eigenvectors.ToArray();
double[][] selectedEigenVectors = eigenVectors.Submatrix(0, eigenVectors.Length - 1, indexes);
this.W = pca.getW().Multiply(selectedEigenVectors);
// Construct projectedTrainingMatrix
this.projectedTrainingSet = new List<projectedTrainingMatrix>();
for (int i = 0; i < trainingSet.Length; i++) {
projectedTrainingMatrix ptm = new projectedTrainingMatrix(this.W.Transpose().Multiply(trainingSet[i].Subtract(pca.meanMatrix).ToArray()), labels[i]);
this.projectedTrainingSet.Add(ptm);
}
this.meanMatrix= pca.meanMatrix;
GC.Collect();
}
void start()
{
DirectoryInfo dirInfo = new DirectoryInfo(@"E:\Work\Work apps\Antony\IFI_DB");
info = dirInfo.GetFiles("*.jpg");
//info.Length
numberofimages = info.Length;
numOfClasses = numberofimages / ImagePerClass;
mymat = new List<double[]>();
List<string> filenames = new List<string>();
for (int i = 0; i < numberofimages; i++)
{
if (info[i].Name.Substring(3, 2) != "07")
{
filenames.Add(info[i].Name.Substring(0, 3));
var img = BitmapFactory.New(1, 1).FromStream(info[i].OpenRead());
if (img.PixelWidth != 100 || img.PixelHeight != 150)
{
img = img.Resize(100, 150, WriteableBitmapExtensions.Interpolation.Bilinear);
}
mymat.Add(helper.imgtomat(img.ToByteArray()));
}
}
GC.Collect();
if (decide == 0)
{
pca = new PCA(mymat.ToArray(), filenames, filenames.Count);
}
else
{
if (File.Exists(@"E:\Work\Work apps\lda pca face rec\outputFile.txt"))
{
ObjectToSerialize objectToSerialize = Serializer.DeSerializeObject(@"E:\Work\Work apps\lda pca face rec\outputFile.txt");
lda = objectToSerialize.Lda;
}
else
{
lda = new LDA(mymat.ToArray(), filenames, filenames.Count);
ObjectToSerialize objectToSerialize = new ObjectToSerialize();
objectToSerialize.Lda = lda ;
Serializer serializer = new Serializer();
serializer.SerializeObject(@"E:\Work\Work apps\lda pca face rec\outputFile.txt", objectToSerialize);
}
}
}
public LDA(double[][] trainingSet, List <string> labels, int numOfComponents)
{
int n = trainingSet.Length; // sample size
HashSet <String> tempSet = new HashSet <String>(labels);
int c = tempSet.Count; // class size
// process in PCA
PCA pca = new PCA(trainingSet, labels, n - c);
// classify
double[][] meanTotal = new double[n - c][];
for (int i = 0; i < n - c; i++)
{
meanTotal[i] = new double[1];
}
Dictionary <String, List <double[]> > dict = new Dictionary <String, List <double[]> >();
List <projectedTrainingMatrix> pcaTrain = pca.getProjectedTrainingSet();
for (int i = 0; i < pcaTrain.Count; i++)
{
String key = pcaTrain[i].label;
meanTotal = meanTotal.Add(pcaTrain[i].matrix);
if (!dict.ContainsKey(key))
{
List <double[]> temp = new List <double[]>();
temp.Add(pcaTrain[i].matrix.Transpose()[0]);
dict.Add(key, temp);
}
else
{
List <double[]> temp = dict[key];
temp.Add(pcaTrain[i].matrix.Transpose()[0]);
dict[key] = temp;
}
}
meanTotal.ToMatrix().Multiply((double)1 / n);
// calculate Sw, Sb
double[][] Sw = new double[n - c][];
double[][] Sb = new double[n - c][];
for (int i = 0; i < n - c; i++)
{
Sw[i] = new double[n - c];
Sb[i] = new double[n - c];
}
List <String> labelSet = dict.Keys.ToList();
foreach (string label in labelSet)
{
List <double[]> tempMatrix = dict[label];
double[][] matrixWithinThatClass = tempMatrix.ToArray();
double[] meanOfCurrentClass = Accord.Statistics.Tools.Mean(matrixWithinThatClass);
for (int i = 0; i < matrixWithinThatClass.Length; i++)
{
double[][] temp1 = matrixWithinThatClass[i].ToArray().Subtract(meanOfCurrentClass.ToArray());
temp1 = temp1.Multiply(temp1.Transpose());
Sw = Sw.Add(temp1);
}
double[][] temp = meanOfCurrentClass.ToArray().Subtract(meanTotal);
temp = temp.Multiply(temp.Transpose()).ToMatrix().Multiply((double)matrixWithinThatClass.Length).ToArray();
Sb = Sb.Add(temp);
}
// calculate the eigenvalues and vectors of Sw^-1 * Sb
double [][] targetForEigen = Sw.Inverse().Multiply(Sb);
var feature = new EigenvalueDecomposition(targetForEigen.ToMatrix());
double[] d = feature.RealEigenvalues;
int[] indexes;
d.StableSort(out indexes);
indexes = indexes.Reverse().ToArray();
indexes = indexes.Submatrix(0, c - 1);
//int[] indexes = getIndexesOfKEigenvalues(d, c - 1);
double[][] eigenVectors = feature.Eigenvectors.ToArray();
double[][] selectedEigenVectors = eigenVectors.Submatrix(0, eigenVectors.Length - 1, indexes);
this.W = pca.getW().Multiply(selectedEigenVectors);
// Construct projectedTrainingMatrix
this.projectedTrainingSet = new List <projectedTrainingMatrix>();
for (int i = 0; i < trainingSet.Length; i++)
{
projectedTrainingMatrix ptm = new projectedTrainingMatrix(this.W.Transpose().Multiply(trainingSet[i].Subtract(pca.meanMatrix).ToArray()), labels[i]);
this.projectedTrainingSet.Add(ptm);
}
this.meanMatrix = pca.meanMatrix;
GC.Collect();
}