/// <summary>
/// Converts the image.
/// </summary>
/// <param name="pathIn">The path in.</param>
/// <param name="pathOut">The path out.</param>
public void ConvertImages(string pathIn, string pathOut)
{
int counter = 0;
var fileList = Directory.GetFiles(@pathIn);
int count = fileList.Length;
var multiplier = 100d / count;
Directory.CreateDirectory(Constants.PRE_FOLDER_BACKGROUND);
Directory.CreateDirectory(Constants.PRE_FOLDER_HUMAN);
foreach (string fileName in fileList)
{
BitmapSource image = new BitmapImage(new Uri(fileName, UriKind.RelativeOrAbsolute));
var preprocessedArray = PreprocessImage(image, Filter.Sobel);
var preprocessedImage = AuxiliaryMethods.BitmapSourceFromArray(preprocessedArray, false);
var resizedImage = ResizeImage(preprocessedImage);
resizedImage.SaveImage($"{pathOut}\\{counter}.png");
double percentage = counter * multiplier;
counter++;
ReportProgress?.Invoke(this, new ProgressArgs {
Percentage = percentage, Message = string.Empty
});
}
}
/// <summary>
/// Initializes a new instance of the <see cref="HogProcessor"/> class.
/// </summary>
public HogProcessor()
{
if (File.Exists(Constants.PATH_SVM))
{
_svm = AuxiliaryMethods.DeserializeSVM(Constants.PATH_SVM);
}
_imageProcessor = new ImageProcessor();
_clusterisationProcessor = new ClusterisationProcessor();
}
/// <summary>
/// Adds the training samples to database.
/// </summary>
/// <param name="pathIn">The path in.</param>
/// <param name="pathOut">The path out.</param>
public void AddTrainingSamplesToDatabase(string pathIn, string pathOut)
{
var samplesHuman = new List <ObjectOfRecognition>();
var samplesBackGround = new List <ObjectOfRecognition>();
int isHuman = pathIn.Contains(Constants.HUMAN_TITLE) ? 1 : 0;
long counter = 0;
var fileList = Directory.GetFiles(pathIn);
int count = fileList.Length;
foreach (string filename in fileList)
{
if (!filename.Contains(".png") && !filename.Contains(".jpg"))
{
continue;
}
var image = new BitmapImage(new Uri(filename, UriKind.RelativeOrAbsolute));
//var bimapImage = new Bitmap(filename);
var byteArray = image.ToByteArray();
var hogVector = ExtractHog(byteArray);
//var hogVector = ExtractHogAccord(bimapImage);
var objOfRec = new ObjectOfRecognition
{
ID = counter,
IsHuman = isHuman,
HOG = hogVector.ToArray()
};
if (pathIn.Contains(Constants.HUMAN_TITLE))
{
samplesHuman.Add(objOfRec);
}
else
{
samplesBackGround.Add(objOfRec);
}
double percentage = counter * 100d / count;
counter++;
ReportProgress?.Invoke(this, new ProgressArgs {
Percentage = percentage, Message = string.Empty
});
}
AuxiliaryMethods.Serialize(pathIn.Contains(Constants.HUMAN_TITLE) ? samplesHuman : samplesBackGround, pathOut);
}
/// <summary>
/// Alls the passes of window.
/// </summary>
/// <param name="src">The source.</param>
/// <param name="step">The step.</param>
/// <returns></returns>
public Task <List <RecognizedObject>[]> AllPassesOfWindow(BitmapSource src, int step)
{
var preprocessedArray = _imageProcessor.PreprocessImage(src, Filter.Sobel);
var preprocessedImage = AuxiliaryMethods.BitmapSourceFromArray(preprocessedArray, false);
int width = Constants.WIDTH;
int height = Constants.HEIGHT;
var taskList = new List <Task <List <RecognizedObject> > >();
while (width < preprocessedImage.Width && height < preprocessedImage.Height)
{
src.Freeze();
taskList.Add(OnePassOfWindow(src, width, height, step));
width = (int)Math.Round(width * 1.5);
height = (int)Math.Round(height * 1.5);
}
return(Task.WhenAll(taskList));
}
/// <summary>
/// Trains the specified true path.
/// </summary>
/// <param name="truePath">The true path.</param>
/// <param name="falsePath">The false path.</param>
public void Train(string truePath, string falsePath)
{
var samples = new List <ObjectOfRecognition>();
samples.AddRange(AuxiliaryMethods.Deserialize <List <ObjectOfRecognition> >(truePath));
samples.AddRange(AuxiliaryMethods.Deserialize <List <ObjectOfRecognition> >(falsePath));
var trainArray = samples.Select(item => item.HOG).ToArray();
var outputArray = samples.Select(item => item.IsHuman).ToArray();
var teacher = new SequentialMinimalOptimization <Gaussian>
{
UseComplexityHeuristic = true,
UseKernelEstimation = true
};
SupportVectorMachine <Gaussian> svm = teacher.Learn(trainArray, outputArray);
var resultLine = svm.Weights;
AuxiliaryMethods.WriteWeight(resultLine, Constants.PATH_WEIGHT);
AuxiliaryMethods.SerializeSVM(svm, Constants.PATH_SVM);
}
