private static Rectangle process(Gray<byte>[,] probabilityMap, Rectangle roi, TermCriteria termCriteria, out CentralMoments centralMoments)
{
Rectangle imageArea = new Rectangle(0, 0, probabilityMap.Width(), probabilityMap.Height());
Rectangle searchWindow = roi;
RawMoments moments = new RawMoments(order: 1);
// Mean shift with fixed number of iterations
int i = 0;
double shift = Byte.MaxValue;
while (termCriteria.ShouldTerminate(i, shift) == false && !searchWindow.IsEmptyArea())
{
// Locate first order moments
moments.Compute(probabilityMap, searchWindow);
int shiftX = (int)(moments.CenterX - searchWindow.Width / 2f);
int shiftY = (int)(moments.CenterY - searchWindow.Height / 2f);
// Shift the mean (centroid)
searchWindow.X += shiftX;
searchWindow.Y += shiftY;
// Keep the search window inside the image
searchWindow.Intersect(imageArea);
shift = System.Math.Abs((double)shiftX) + System.Math.Abs((double)shiftY); //for term criteria only
i++;
}
if (searchWindow.IsEmptyArea() == false)
{
// Locate second order moments and perform final shift
moments.Order = 2;
moments.Compute(probabilityMap, searchWindow);
searchWindow.X += (int)(moments.CenterX - searchWindow.Width / 2f);
searchWindow.Y += (int)(moments.CenterY - searchWindow.Height / 2f);
// Keep the search window inside the image
searchWindow.Intersect(imageArea);
}
centralMoments = new CentralMoments(moments); // moments to be used by camshift
return searchWindow;
}
private unsafe static void goodFeaturesToTrack(Gray <float>[,] integralDxx, Gray <float>[,] integralDxy, Gray <float>[,] integralDyy,
int winSize, float minEigValue, Gray <float>[,] strengthImg)
{
minEigValue = System.Math.Max(1E-3f, minEigValue);
int normFactor = winSize * winSize * 255;
int maxCol = integralDxx.Width() - winSize;
int maxRow = integralDxx.Height() - winSize;
for (int row = 0; row < maxRow; row++)
{
for (int col = 0; col < maxCol; col++)
{
var Dxx = integralDxx.GetSum(col, row, winSize, winSize);
var Dxy = integralDxy.GetSum(col, row, winSize, winSize);
var Dyy = integralDyy.GetSum(col, row, winSize, winSize);
var eigenVal = calcMinEigenVal(Dxx, Dxy, Dyy);
eigenVal /= normFactor;
if (eigenVal > minEigValue)
{
strengthImg[winSize / 2 + row, winSize / 2 + col] = eigenVal;
}
}
}
}
private unsafe static void goodFeaturesToTrack(Gray<float>[,] integralDxx, Gray<float>[,] integralDxy, Gray<float>[,] integralDyy,
int winSize, float minEigValue, Gray<float>[,] strengthImg)
{
minEigValue = System.Math.Max(1E-3f, minEigValue);
int normFactor = winSize * winSize * 255;
int maxCol = integralDxx.Width() - winSize;
int maxRow = integralDxx.Height() - winSize;
for (int row = 0; row < maxRow; row++)
{
for (int col = 0; col < maxCol; col++)
{
var Dxx = integralDxx.GetSum(col, row, winSize, winSize);
var Dxy = integralDxy.GetSum(col, row, winSize, winSize);
var Dyy = integralDyy.GetSum(col, row, winSize, winSize);
var eigenVal = calcMinEigenVal(Dxx, Dxy, Dyy);
eigenVal /= normFactor;
if (eigenVal > minEigValue)
{
strengthImg[winSize / 2 + row, winSize / 2 + col] = eigenVal;
}
}
}
}
private static ComplexF[,] prepareImage(Gray <float>[,] image, int biggestKernelWidth, int biggestKernelHeight,
ConvolutionBorder options,
out int fillX, out int fillY)
{
int FFTNumOfCols = (int)System.Math.Pow(2.0, System.Math.Ceiling(System.Math.Log(biggestKernelWidth + image.Width(), 2.0)));
int FFTNumOfRows = (int)System.Math.Pow(2.0, System.Math.Ceiling(System.Math.Log(biggestKernelHeight + image.Height(), 2.0)));
fillX = System.Math.Min(image.Width(), biggestKernelWidth / 2);
fillY = System.Math.Min(image.Height(), biggestKernelHeight / 2);
var paddedImage = new Gray <float> [FFTNumOfRows, FFTNumOfCols];
//center
image.CopyTo(paddedImage, new Point(fillX, fillY));
if (options == ConvolutionBorder.BorderMirror)
{
mirrorBorders(image, paddedImage, fillX, fillY);
}
var paddedImageCmplx = paddedImage.ToComplex();
paddedImageCmplx.FFT(FourierTransform.Direction.Forward, true);
return(paddedImageCmplx);
}
private static Rectangle process(Gray <byte>[,] probabilityMap, Rectangle roi, TermCriteria termCriteria, out CentralMoments centralMoments)
{
Rectangle imageArea = new Rectangle(0, 0, probabilityMap.Width(), probabilityMap.Height());
Rectangle searchWindow = roi;
RawMoments moments = new RawMoments(order: 1);
// Mean shift with fixed number of iterations
int i = 0;
double shift = Byte.MaxValue;
while (termCriteria.ShouldTerminate(i, shift) == false && !searchWindow.IsEmptyArea())
{
// Locate first order moments
moments.Compute(probabilityMap, searchWindow);
int shiftX = (int)(moments.CenterX - searchWindow.Width / 2f);
int shiftY = (int)(moments.CenterY - searchWindow.Height / 2f);
// Shift the mean (centroid)
searchWindow.X += shiftX;
searchWindow.Y += shiftY;
// Keep the search window inside the image
searchWindow.Intersect(imageArea);
shift = System.Math.Abs((double)shiftX) + System.Math.Abs((double)shiftY); //for term criteria only
i++;
}
if (searchWindow.IsEmptyArea() == false)
{
// Locate second order moments and perform final shift
moments.Order = 2;
moments.Compute(probabilityMap, searchWindow);
searchWindow.X += (int)(moments.CenterX - searchWindow.Width / 2f);
searchWindow.Y += (int)(moments.CenterY - searchWindow.Height / 2f);
// Keep the search window inside the image
searchWindow.Intersect(imageArea);
}
centralMoments = new CentralMoments(moments); // moments to be used by camshift
return(searchWindow);
}
/// <summary>
/// Computes gradient orientations from the color image. Orientation from the channel which has the maximum gradient magnitude is taken as the orientation for a location.
/// </summary>
/// <param name="frame">Image.</param>
/// <param name="magnitudeSqrImage">Squared magnitude image.</param>
/// <param name="minValidMagnitude">Minimal valid magnitude.</param>
/// <returns>Orientation image (angles are in degrees).</returns>
public static unsafe Gray<int>[,] Compute(Gray<byte>[,] frame, out Gray<int>[,] magnitudeSqrImage, int minValidMagnitude)
{
var minSqrMagnitude = minValidMagnitude * minValidMagnitude;
var orientationImage = new Gray<int>[frame.Height(), frame.Width()];
var _magnitudeSqrImage = orientationImage.CopyBlank();
using (var uFrame = frame.Lock())
{
ParallelLauncher.Launch(thread =>
{
computeGray(thread, (byte*)uFrame.ImageData, uFrame.Stride, orientationImage, _magnitudeSqrImage, minSqrMagnitude);
},
frame.Width() - 2 * kernelRadius, frame.Height() - 2 * kernelRadius);
}
magnitudeSqrImage = _magnitudeSqrImage;
return orientationImage;
}
private static Gray<byte>[,] QuantizeOrientations(Gray<int>[,] orientDegImg)
{
var quantizedUnfilteredOrient = new Gray<byte>[orientDegImg.Height(), orientDegImg.Width()];
using (var uOrientDegImg = orientDegImg.Lock())
using (var uQuantizedUnfilteredOrient = quantizedUnfilteredOrient.Lock())
{
int* orientDegImgPtr = (int*)uOrientDegImg.ImageData;
byte* qOrinetUnfilteredPtr = (byte*)uQuantizedUnfilteredOrient.ImageData;
int qOrinetUnfilteredStride = uQuantizedUnfilteredOrient.Stride;
int imgWidth = uOrientDegImg.Width;
int imgHeight = uOrientDegImg.Height;
for (int j = 0; j < imgHeight; j++)
{
for (int i = 0; i < imgWidth; i++)
{
int angle = orientDegImgPtr[i];
qOrinetUnfilteredPtr[i] = AngleQuantizationTable[angle]; //[0-360] -> [...] -> [0-7] (for mapping see "CalculateAngleQuantizationTable()")
}
orientDegImgPtr += imgWidth; //<Gray<int>> is always alligned
qOrinetUnfilteredPtr += qOrinetUnfilteredStride;
}
}
//quantizedUnfilteredOrient.Mul(36).Save("quantizedUnfilteredImg.bmp");
return quantizedUnfilteredOrient;
}
/// <summary>
/// Creates linear response maps.
/// </summary>
/// <param name="orientationDegImg">Orientation image (in degrees).</param>
/// <param name="neigborhood">Spread neighborhood size.</param>
public LinearizedMaps(Gray<int>[,] orientationDegImg, int neigborhood)
{
this.NeigborhoodSize = neigborhood;
this.ImageSize = orientationDegImg.Size();
this.LinearMapSize = new Size(orientationDegImg.Width() / neigborhood, orientationDegImg.Height() / neigborhood);
this.ImageValidSize = new Size(this.LinearMapSize.Width * neigborhood, this.LinearMapSize.Height * neigborhood);
this.LinearMaps = calculate(orientationDegImg);
}
