public void TestGetFeatureValue1()
{
double eps = 0.000001;
Feature feature1 = new Feature(FeatureType.type1, DebugInfo2.x, DebugInfo2.y, DebugInfo2.w, DebugInfo2.h);
Feature feature2 = new Feature(FeatureType.type2, DebugInfo2.x, DebugInfo2.y, DebugInfo2.w, DebugInfo2.h);
Feature feature3 = new Feature(FeatureType.type3, DebugInfo2.x, DebugInfo2.y, DebugInfo2.w, DebugInfo2.h);
Feature feature4 = new Feature(FeatureType.type4, DebugInfo2.x, DebugInfo2.y, DebugInfo2.w, DebugInfo2.h);
double f1 = Feature.getFeatureValue(DebugInfo1.ii_mm, feature1);
double f2 = Feature.getFeatureValue(DebugInfo1.ii_mm, feature2);
double f3 = Feature.getFeatureValue(DebugInfo1.ii_mm, feature3);
double f4 = Feature.getFeatureValue(DebugInfo1.ii_mm, feature4);
double f1ref = DebugInfo1.im.getSubMatrix(DebugInfo2.x, DebugInfo2.y, DebugInfo2.w, DebugInfo2.h).getSum() -
DebugInfo1.im.getSubMatrix(DebugInfo2.x, DebugInfo2.y + DebugInfo2.h, DebugInfo2.w, DebugInfo2.h).getSum();
double f2ref = DebugInfo1.im.getSubMatrix(DebugInfo2.x + DebugInfo2.w, DebugInfo2.y, DebugInfo2.w, DebugInfo2.h).getSum() -
DebugInfo1.im.getSubMatrix(DebugInfo2.x, DebugInfo2.y, DebugInfo2.w, DebugInfo2.h).getSum();
double f3ref = DebugInfo1.im.getSubMatrix(DebugInfo2.x + DebugInfo2.w, DebugInfo2.y, DebugInfo2.w, DebugInfo2.h).getSum() -
DebugInfo1.im.getSubMatrix(DebugInfo2.x, DebugInfo2.y, DebugInfo2.w, DebugInfo2.h).getSum() -
DebugInfo1.im.getSubMatrix(DebugInfo2.x + 2 * DebugInfo2.w, DebugInfo2.y, DebugInfo2.w, DebugInfo2.h).getSum();
double f4ref = DebugInfo1.im.getSubMatrix(DebugInfo2.x + DebugInfo2.w, DebugInfo2.y, DebugInfo2.w, DebugInfo2.h).getSum() +
DebugInfo1.im.getSubMatrix(DebugInfo2.x, DebugInfo2.y + DebugInfo2.h, DebugInfo2.w, DebugInfo2.h).getSum() -
DebugInfo1.im.getSubMatrix(DebugInfo2.x, DebugInfo2.y, DebugInfo2.w, DebugInfo2.h).getSum() -
DebugInfo1.im.getSubMatrix(DebugInfo2.x + DebugInfo2.w, DebugInfo2.y + DebugInfo2.h, DebugInfo2.w, DebugInfo2.h).getSum();
Assert.IsTrue(Math.Abs(f1 - f1ref) < eps);
Assert.IsTrue(Math.Abs(f2 - f2ref) < eps);
Assert.IsTrue(Math.Abs(f3 - f3ref) < eps);
Assert.IsTrue(Math.Abs(f4 - f4ref) < eps);
}
public static Matrix getAllFeatureValues(Matrix[] integral_images, Feature[] allFeatures)
{
Matrix result = new Matrix(integral_images.Length, allFeatures.Length);
for (int i = 1; i <= allFeatures.Length; i++)
{
result.setRow(i, getFeatureValues(integral_images, allFeatures[i-1]));
}
return result;
}
public static AdaBoostRespons[] executeAdaBoost(Matrix[] integralImages, Feature[] allFeatures, Matrix isFaceList, int nrNegative, int nrWeakClassifiers)
{
FeatureType[] featureTypes =
(from feature in allFeatures
select feature.type).Distinct().ToArray();
InitiateFeatureValues(integralImages, allFeatures, featureTypes);
AdaBoostRespons[] adaBoostRespons = new AdaBoostRespons[nrWeakClassifiers];
Matrix weights = getInitializedWeights(isFaceList.nr_vals - nrNegative, nrNegative);
for (int t = 0; t < nrWeakClassifiers; t++)
{
weights = weights / weights.getSum();
FeatureIndex bestFeatureIndex = new FeatureIndex(FeatureType.type1, int.MinValue);
WeakClassifier bestClassifier = new WeakClassifier(int.MaxValue, double.MaxValue, double.MaxValue);
foreach (FeatureType featureType in featureTypes)
{
Matrix allFeatureValuesOfSameType = Matrix.DeserializeFromXML("FeatureValues_" + featureType.ToString() + ".xml");
for (int j = 1; j <= allFeatureValuesOfSameType.nr_rows; j++)
{
WeakClassifier weakClassifier = learnWeakClassifier(allFeatureValuesOfSameType, weights, isFaceList, j);
if (weakClassifier.error < bestClassifier.error)
{
bestClassifier = weakClassifier;
bestFeatureIndex.featureType = featureType;
bestFeatureIndex.j = j;
}
}
}
double beta = bestClassifier.error / (1 - bestClassifier.error);
double alpha = Math.Log(1 / beta);
Matrix featureValuesOfSameType = Matrix.DeserializeFromXML("FeatureValues_" + bestFeatureIndex.featureType.ToString() + ".xml");
for (int i = 1; i <= featureValuesOfSameType.nr_cols; i++)
{
int classification = getWeakClassification(featureValuesOfSameType[i, bestFeatureIndex.j], bestClassifier.parity, bestClassifier.threshold);
weights[i] = Math.Abs(classification - isFaceList[i]) == 0 ? weights[i] * beta : weights[i];
}
adaBoostRespons[t] = new AdaBoostRespons(bestClassifier, alpha, bestFeatureIndex);
}
return adaBoostRespons;
}
public void TestGetFeatureValue2()
{
double eps = 0.000001;
Feature feature1 = new Feature(FeatureType.type1, DebugInfo2.x, DebugInfo2.y, DebugInfo2.w, DebugInfo2.h);
Feature feature2 = new Feature(FeatureType.type2, DebugInfo2.x, DebugInfo2.y, DebugInfo2.w, DebugInfo2.h);
Feature feature3 = new Feature(FeatureType.type3, DebugInfo2.x, DebugInfo2.y, DebugInfo2.w, DebugInfo2.h);
Feature feature4 = new Feature(FeatureType.type4, DebugInfo2.x, DebugInfo2.y, DebugInfo2.w, DebugInfo2.h);
double f1 = Feature.getFeatureValue(DebugInfo1.ii_mm, feature1);
double f2 = Feature.getFeatureValue(DebugInfo1.ii_mm, feature2);
double f3 = Feature.getFeatureValue(DebugInfo1.ii_mm, feature3);
double f4 = Feature.getFeatureValue(DebugInfo1.ii_mm, feature4);
Assert.IsTrue(Math.Abs(DebugInfo2.f1 - f1) < eps);
Assert.IsTrue(Math.Abs(DebugInfo2.f2 - f2) < eps);
Assert.IsTrue(Math.Abs(DebugInfo2.f3 - f3) < eps);
Assert.IsTrue(Math.Abs(DebugInfo2.f4 - f4) < eps);
}
public Classifier(double[] alphas, double[] thetas, Feature[] features, Feature[] all_features)
{
this.alphas = alphas;
this.thetas = thetas;
this.features = features;
this.all_features = all_features;
}
private static void InitiateFeatureValues(Matrix[] integralImages, Feature[] allFeatures, FeatureType[] featureTypes)
{
foreach (FeatureType featureType in featureTypes)
{
Feature[] allFeaturesOfSameType = (from feature in allFeatures
where feature.type.Equals(featureType)
select feature).ToArray();
Matrix allFeatureValuesOfSameType = Feature.getAllFeatureValues(integralImages, allFeaturesOfSameType);
allFeatureValuesOfSameType.saveToXML("FeatureValues_" + featureType.ToString());
}
}
/// CHECKED
/// <summary>
///
/// </summary>
/// <param name="feature"></param>
/// <param name="featureWidth"></param>
/// <param name="featureHeight"></param>
/// <param name="filename"></param>
public static void saveFeatureImage(Feature feature, int featureWidth, int featureHeight, string filename)
{
int k = 5;
int x = feature.x * k;
int y = feature.y * k;
int w = feature.w * k;
int h = feature.h * k;
int width = featureWidth * k;
int height = featureHeight * k;
Bitmap bitMapImage = new Bitmap(width, height);
switch (feature.type)
{
case FeatureType.type1:
fillRectangle(ref bitMapImage, x, y, w, h, Color.Red);
fillRectangle(ref bitMapImage, x, y + h, w, h, Color.Green);
break;
case FeatureType.type2:
fillRectangle(ref bitMapImage, x, y, w, h, Color.Green);
fillRectangle(ref bitMapImage, x + w, y, w, h, Color.Red);
break;
case FeatureType.type3:
fillRectangle(ref bitMapImage, x, y, w, h, Color.Green);
fillRectangle(ref bitMapImage, x + w, y, w, h, Color.Red);
fillRectangle(ref bitMapImage, x + w + w, y, w, h, Color.Green);
break;
case FeatureType.type4:
fillRectangle(ref bitMapImage, x, y, w, h, Color.Green);
fillRectangle(ref bitMapImage, x + w, y, w, h, Color.Red);
fillRectangle(ref bitMapImage, x, y + h, w, h, Color.Red);
fillRectangle(ref bitMapImage, x + w, y + h, w, h, Color.Green);
break;
}
EncoderParameters encoderParameters = new EncoderParameters(1);
encoderParameters.Param[0] = new EncoderParameter(System.Drawing.Imaging.Encoder.Quality, 100L);
bitMapImage.Save(filename, GetEncoder(ImageFormat.Jpeg), encoderParameters);
}
/// CHECKED
/// <summary>
///
/// </summary>
/// <param name="integral_images"></param>
/// <param name="feature"></param>
/// <returns></returns>
public static Matrix getFeatureValues(Matrix[] integral_images, Feature feature)
{
double[] values = (from integral_image in integral_images
select getFeatureValue(integral_image, feature)).ToArray();
return new Matrix(values.Length, 1, values);
}
/// CHECKED
/// <summary>
///
/// </summary>
/// <param name="integral_image"></param>
/// <param name="feature"></param>
/// <returns></returns>
public static double getFeatureValue(Matrix integral_image, Feature feature)
{
double val = 0;
int x = feature.x;
int y = feature.y;
int w = feature.w;
int h = feature.h;
switch (feature.type)
{
case FeatureType.type1:
val = getBoxSum(integral_image, x, y, w, h);
val -= getBoxSum(integral_image, x, y + h, w, h);
break;
case FeatureType.type2:
val = getBoxSum(integral_image, x + w, y, w, h);
val -= getBoxSum(integral_image, x, y, w, h);
break;
case FeatureType.type3:
val = getBoxSum(integral_image, x + w, y, w, h);
val -= getBoxSum(integral_image, x, y, w, h);
val -= getBoxSum(integral_image, x + 2 * w, y, w, h);
break;
case FeatureType.type4:
val = getBoxSum(integral_image, x + w, y, w, h);
val += getBoxSum(integral_image, x, y + h, w, h);
val -= getBoxSum(integral_image, x, y, w, h);
val -= getBoxSum(integral_image, x + w, y + h, w, h);
break;
}
return val;
}
public void TestSaveFeatureImage()
{
Feature feature1 = new Feature(FeatureType.type1, 2, 2, 3, 4);
Feature.saveFeatureImage(feature1, 19, 19, "test1.jpg");
Feature feature2 = new Feature(FeatureType.type4, 5, 5, 5, 5);
Feature.saveFeatureImage(feature2, 19, 19, "test2.jpg");
Assert.IsFalse(false);
}