public ZagImage <int> SmoothAdaptive(ZagImage <int> origQuantGradient)
{
var quantGradient = new ZagImage <int>(origQuantGradient.Width, origQuantGradient.Height, 1);
var smoothQuantGradient = new ZagImage <int>(origQuantGradient.Width, origQuantGradient.Height, 1);
//quantGradient.SetZero();
smoothQuantGradient.SetValue(127);
for (int y = 0; y < quantGradient.Height; y++)
{
for (int x = 0; x < quantGradient.Width; x++)
{
quantGradient.Data[y * quantGradient.Stride + x] = (int)Math.Round((origQuantGradient.Data[y * origQuantGradient.Stride + x] + 179) * 0.7111111111111);
}
}
int k = 1;
float factor = 3;
double weights = 0;
double weightPixels = 0;
float Gx = 0, Gy = 0;
for (int y = k + 1; y < quantGradient.Height - k - 1; y++)
{
for (int x = k + 1; x < quantGradient.Width - k - 1; x++)
{
if (quantGradient.Data[y * quantGradient.Stride + x] != 127)
{
weightPixels = 0;
weights = 0;
Gx = 0;
Gy = 0;
for (int j = -k; j <= k; j++)
{
for (int i = -k, X = 0; i <= k; i++, X++)
{
Gx = (quantGradient.Data[(y + i) * quantGradient.Stride + x + j + 1] - quantGradient.Data[(y + i) * quantGradient.Stride + x + j - 1]) / 2f;
Gy = (quantGradient.Data[(y + i + 1) * quantGradient.Stride + x + j] - quantGradient.Data[(y + i - 1) * quantGradient.Stride + x + j]) / 2f;
double w = Math.Exp(-1 * (Gx * Gx + Gy * Gy) / (2 * factor * factor));
weights += w;
weightPixels += w * (quantGradient.Data[(y + i) * quantGradient.Stride + x + j]);
}
}
smoothQuantGradient.Data[y * smoothQuantGradient.Stride + x] = Math.Abs(weights) < double.Epsilon ? 0 : (int)Math.Round(weightPixels / weights);
}
}
}
for (int y = 0; y < smoothQuantGradient.Height; y++)
{
for (int x = 0; x < smoothQuantGradient.Width; x++)
{
smoothQuantGradient.Data[y * smoothQuantGradient.Stride + x] = (int)Math.Round(1.40625 * smoothQuantGradient.Data[y * smoothQuantGradient.Stride + x] - 179);
}
}
return(smoothQuantGradient);
}
public void CalcFeatures(ZagImage <byte> origImg, ZagImage <int> gradientImg, ZagImage <float> Magnitude)
{
var newGradient = firstGrandientWithoutThreshold(origImg);
int windowR = 0;
DynamicWindow = false;
for (int i = 0; i < myDsLPs.Count(); i++)
{
float[] features = null;
myDsLPs[i].WindowSize = (DynamicWindow) ? CalcWindowSizeAverage(myDsLPs[i], origImg, this.WindowSize, 4, this.N) : this.WindowSize;
windowR = myDsLPs[i].WindowSize / N;
features = GradientDescriptorNxN(myDsLPs[i], newGradient.Item1, newGradient.Item2, windowR, N);
NormalizeDescriptorAndThresholding(ref features, 0.2f);
myDsLPs[i].Descriptor = features;
}
}
public ZagImage <int> QuantizeFromTables(ZagImage <int> gradient)
{
var quantGradient = new ZagImage <int>(gradient.Width, gradient.Height, 1);
var maxQuantizationLevel = gradientLevels[(int)this.QuantizationLevels].Length;
for (int y = 0; y < gradient.Height; y++)
{
for (int x = 0; x < gradient.Width; x++)
{
if (gradient.Data[y * gradient.Stride + x] != 0)
{
if (gradient.Data[y * gradient.Stride + x] > gradientLevels[(int)this.QuantizationLevels][maxQuantizationLevel - 1])
{
quantGradient.Data[y * quantGradient.Stride + x] = maxQuantizationLevel;
}
else
{
for (int q = 0; q < maxQuantizationLevel; q++)
{
if (gradient.Data[y * gradient.Stride + x] < gradientLevels[(int)this.QuantizationLevels][q])
{
quantGradient.Data[y * quantGradient.Stride + x] = q + 1;
break;
}
}
}
}
}
}
return(quantGradient);
}
public List <DoLF.DsLPoints> getCenterofGravityPoints(ZagImage <byte> img, int MinCCSize)
{
var myblobCounter = new BlobCounter();
var myblobs = myblobCounter.GetObjectsWithoutArray(img).Where(x => x.Width >= MinCCSize && x.Height >= MinCCSize);
//if (myblobs.Count() > 1)
//{
// // Calculate Mean Area Size
// double average = myblobs.Average(x => x.Area);
// //Calculate Average Absolute Deviation
// double aad = myblobs.Select(x => Math.Abs(x.Area - average)).Average();
// myblobs = myblobs.Where(x => x.Area >= average - aad && x.Area <= average + aad).ToArray();
//}
var mypoints = new List <DoLF.DsLPoints>();
foreach (var b in myblobs)
{
mypoints.Add(new DoLF.DsLPoints((int)Math.Round(b.CenterOfGravityX), (int)Math.Round(b.CenterOfGravityY)));
}
return(mypoints);
}
/// <summary>
/// Get the DSLP Local Points
/// </summary>
/// <param name="Image">an 24bit RGB Image</param>
/// <param name="Width"></param>
/// <param name="Height"></param>
/// <returns></returns>
public DsLPoints[] GetDSLPoints(byte[] Image, int Width, int Height, int ImageDepth, int WindowSize = 68, int N = 4, QuantizationLevelsNum QuantizationLevels = QuantizationLevelsNum.Four, int MinCCSize = 4, bool DynamicWindow = true)
{
this.WindowSize = WindowSize;
this.N = N;
this.QuantizationLevels = QuantizationLevels;
this.MinCCSize = MinCCSize;
this.DynamicWindow = DynamicWindow;
var NewPixels = new byte[3 * Width * Height];
int cstride = 3 * Width;
int stride = ImageDepth * Width;
switch (ImageDepth)
{
case 1:
for (int y = 0; y < Height; y++)
{
for (int x = 0; x < Width; x++)
{
NewPixels[y * cstride + 3 * x]
= NewPixels[y * cstride + 3 * x + 1]
= NewPixels[y * cstride + 3 * x + 2]
= Image[y * stride + ImageDepth * x];
}
}
break;
case 4:
for (int y = 0; y < Height; y++)
{
for (int x = 0; x < Width; x++)
{
NewPixels[y * cstride + 3 * x] = Image[y * stride + ImageDepth * x];
NewPixels[y * cstride + 3 * x + 1] = Image[y * stride + ImageDepth * x + 1];
NewPixels[y * cstride + 3 * x + 2] = Image[y * stride + ImageDepth * x + 2];
}
}
break;
case 3:
NewPixels = Image;
break;
default:
throw new Exception("Not Supported Depth");
}
var myImage = new ZagImage <byte>(NewPixels, Width, Height);
var myDSLP = new DsLPsDetectorv2wFeatures
{
N = this.N,
WindowSize = this.WindowSize,
QuantizationLevels = this.QuantizationLevels,
MinCCSize = this.MinCCSize,
DynamicWindow = this.DynamicWindow
};
myDSLP.CalcDoLFs(myImage);
return(myDSLP.myDsLPs.Count() > 0 ? myDSLP.myDsLPs.ToArray() : new DsLPoints[0]);
}
private int FindThreshold(ZagImage <int> SourceImage)
{
int width = SourceImage.Width;
int height = SourceImage.Height;
var histog = new int[256];
int k, threshold = 1;
for (int x = 0; x < width; x++)
{
for (int y = 0; y < height; y++)
{
if (SourceImage.Data[y * SourceImage.Stride + x] != 0)
{
histog[Math.Abs(SourceImage.Data[y * SourceImage.Stride + x])]++;
}
}
}
double sum = 0, csum = 0, fmax = -1, sb = 0, m1 = 0, m2 = 0;
int n = 0, n1 = 0, n2 = 0;
for (k = 0; k < 256; k++)
{
sum += (double)k * histog[k];
n += histog[k];
}
for (k = 0; k < 256; k++)
{
n1 += histog[k];
if (n1 == 0)
{
continue;
}
n2 = n - n1;
if (n == 0)
{
break;
}
csum += (double)k * histog[k];
m1 = csum / n1;
m2 = (sum - csum) / n2;
sb = (double)n1 * n2 * (m1 - m2) * (m1 - m2);
if (sb > fmax)
{
fmax = sb;
threshold = k;
}
}
if (threshold == 0)
{
threshold = 127;
}
return(threshold);
}
public void CalcDoLFs(ZagImage <byte> Img)
{
if (Img.Width > 5 && Img.Height > 5)
{
ZagImage <byte> origImg = GetGrayscaleRMY(Img);
ZagImage <byte> mySmoothMedianImg = SmoothMedian(origImg);
mySmoothMedianImg = SmoothGaussian(mySmoothMedianImg);
Tuple <ZagImage <int>, ZagImage <float> > results = firstGrandient(mySmoothMedianImg);
ZagImage <int> orientationMatrix = results.Item1;
ZagImage <float> magnitudeMatrix = results.Item2;
// smoothing
orientationMatrix = SmoothMedian(orientationMatrix);
orientationMatrix = SmoothAdaptive(orientationMatrix);
ZagImage <int> quantMatrix = QuantizeFromTables(orientationMatrix);
ZagImage <int>[] levels = getLevels(quantMatrix, (int)this.QuantizationLevels);
ZagImage <byte>[] levelImg = levels.Select(x => getLevelImage(x)).ToArray();
var totalPoints = new List <DoLF.DsLPoints>();
for (int i = 0; i < levelImg.Length; i++)
{
List <DoLF.DsLPoints> points = getCenterofGravityPoints(levelImg[i], this.MinCCSize);
totalPoints.AddRange(points);
}
myDsLPs = totalPoints;
for (int i = 0; i < myDsLPs.Count(); i++)
{
DoLF.DsLPoints p = myDsLPs[i];
//p.Gradient = quantMatrix.Data[(int)(p.Y * quantMatrix.Stride + p.X)];
}
CalcFeatures(origImg, quantMatrix, magnitudeMatrix);
}
else
{
myDsLPs = new List <DoLF.DsLPoints>();
}
}
private void Invert(ref ZagImage <byte> SourceImage)
{
int stride = 3 * SourceImage.Width;
int x, y;
// for each line
for (y = 0; y < SourceImage.Height; y++)
{
// for each pixel
for (x = 0; x < SourceImage.Width; x++)
{
SourceImage.Data[3 * x + y * stride] = (byte)(255 - SourceImage.Data[3 * x + y * stride]);
SourceImage.Data[3 * x + y * stride + 1] = (byte)(255 - SourceImage.Data[3 * x + y * stride + 1]);
SourceImage.Data[3 * x + y * stride + 2] = (byte)(255 - SourceImage.Data[3 * x + y * stride + 2]);
}
}
}
private Tuple <ZagImage <int>, ZagImage <int> > getGradients(ZagImage <byte> myImage)
{
var gradientX = new ZagImage <int>(myImage.Width, myImage.Height, 1);
var gradientY = new ZagImage <int>(myImage.Width, myImage.Height, 1);
for (int y = 1; y < myImage.Height - 1; y++)
{
for (int x = 1; x < myImage.Width - 1; x++)
{
int gX = myImage.Data[(x + 1) + y * myImage.Stride] - myImage.Data[(x - 1) + y * myImage.Stride];
int gY = myImage.Data[x + (y + 1) * myImage.Stride] - myImage.Data[x + (y - 1) * myImage.Stride];
gradientX.Data[x + y * myImage.Stride] = gX;
gradientY.Data[x + y * myImage.Stride] = gY;
}
}
return(new Tuple <ZagImage <int>, ZagImage <int> >(gradientX, gradientY));
}
public ZagImage <int>[] getLevels(ZagImage <int> gradient, int qLevels)
{
var level = new ZagImage <int> [qLevels];
for (int i = 0; i < qLevels; i++)
{
level[i] = new ZagImage <int>(gradient.Width, gradient.Height, 1);
}
for (int y = 0; y < gradient.Height; y++)
{
for (int x = 0; x < gradient.Width; x++)
{
if (gradient.Data[y * gradient.Stride + x] != 0)
{
level[gradient.Data[y * gradient.Stride + x] - 1].Data[y * gradient.Stride + x] = 255;
}
}
}
return(level);
}
public Tuple <ZagImage <int>, ZagImage <float> > firstGrandient(ZagImage <byte> origImg)
{
//var mygradientX = new ConvMatrix(new int[] { 0, 0, 0, -1, 0, 1, 0, 0, 0 }, 3);
//var mygradientY = new ConvMatrix(new int[] { 0, -1, 0, 0, 0, 0, 0, 1, 0 }, 3);
//var imageX = Convolution(origImg, mygradientX);
//var imageY = Convolution(origImg, mygradientY);
var gradients = getGradients(origImg);
var imageX = gradients.Item1;
var imageY = gradients.Item2;
int thrX = FindThreshold(imageX) / 2;
int thrY = FindThreshold(imageY) / 2;
var magnitude = new ZagImage <float>(origImg.Width, origImg.Height, 1);
var orientation = new ZagImage <int>(origImg.Width, origImg.Height, 1);
for (int y = 0; y < orientation.Height; y++)
{
for (int x = 0; x < orientation.Width; x++)
{
if (Math.Abs(imageX.Data[x + y * imageX.Stride]) < thrX)
{
imageX.Data[x + y * imageX.Stride] = 0;
}
if (Math.Abs(imageY.Data[x + y * imageY.Stride]) < thrY)
{
imageY.Data[x + y * imageY.Stride] = 0;
}
orientation.Data[x + y * orientation.Stride] =
(int)Math.Round(180 * Math.Atan2(imageY.Data[y * imageY.Stride + x], imageX.Data[y * imageX.Stride + x]) / Math.PI);
magnitude.Data[y * magnitude.Stride + x] =
(float)
Math.Sqrt(imageX.Data[y * imageX.Stride + x] * imageX.Data[y * imageX.Stride + x] +
imageY.Data[y * imageY.Stride + x] * imageY.Data[y * imageY.Stride + x]);
}
}
return(new Tuple <ZagImage <int>, ZagImage <float> >(orientation, magnitude));
}
public ZagImage <byte> SmoothMedian(ZagImage <byte> quantGradient, int k = 1)
{
ZagImage <byte> smoothQuantGradient = quantGradient.Copy();
for (int y = k; y < quantGradient.Height - k; y++)
{
for (int x = k; x < quantGradient.Width - k; x++)
{
var points = new List <byte>();
for (int i = -k; i <= k; i++)
{
for (int j = -k; j <= k; j++)
{
points.Add(quantGradient.Data[(y + i) * quantGradient.Stride + (x + j)]);
}
}
smoothQuantGradient.Data[y * smoothQuantGradient.Stride + x] = points.OrderBy(a => a).ToArray()[points.Count() >> 1];
}
}
return(smoothQuantGradient);
}
private ZagImage <byte> GetGrayscaleRMY(ZagImage <byte> SourceImage)
{
var gray = new ZagImage <byte>(SourceImage.Width, SourceImage.Height, 1);
double cr = 0.5, cg = 0.419, cb = 0.081;
// get width and height
int x, y;
// for each line
for (y = 0; y < SourceImage.Height; y++)
{
// for each pixel
for (x = 0; x < SourceImage.Width; x++)
{
var greyValue = (byte)(cb * SourceImage.Data[SourceImage.Depth * x + y * SourceImage.Stride] + cg * SourceImage.Data[SourceImage.Depth * x + 1 + y * SourceImage.Stride] + cr * SourceImage.Data[SourceImage.Depth * x + 2 + y * SourceImage.Stride]);
gray.Data[gray.Depth * x + y * gray.Stride] = greyValue;
}
}
return(gray);
}
private void SetDebugImage <T>(ZagImage <T> Test)
{
if (typeof(T) == typeof(byte))
{
DoLF.TestImageData = new byte[Test.Data.Length];
Test.Data.CopyTo(DoLF.TestImageData, 0);
}
else if (typeof(T) == typeof(int))
{
DoLF.TestMatrixData = new int[Test.Data.Length];
Test.Data.CopyTo(DoLF.TestMatrixData, 0);
}
else if (typeof(T) == typeof(float))
{
var TestInt = Test.Data.Select(x => Convert.ToInt32(x)).ToArray();
DoLF.TestMatrixData = new int[Test.Data.Length];
TestInt.CopyTo(DoLF.TestMatrixData, 0);
}
DoLF.TestImageWidth = Test.Width;
DoLF.TestImageHeight = Test.Height;
DoLF.Depth = Test.Depth;
}
public Tuple <ZagImage <int>, ZagImage <float> > firstGrandientWithoutThreshold(ZagImage <byte> origImg)
{
var gradients = getGradients(origImg);
var imageX = gradients.Item1;
var imageY = gradients.Item2;
var magnitude = new ZagImage <float>(origImg.Width, origImg.Height);
var orientation = new ZagImage <int>(origImg.Width, origImg.Height);
for (int y = 0; y < orientation.Height; y++)
{
for (int x = 0; x < orientation.Width; x++)
{
orientation.Data[x + y * orientation.Stride] =
(int)Math.Round(180 * Math.Atan2(imageY.Data[y * imageY.Stride + x], imageX.Data[y * imageX.Stride + x]) / Math.PI);
magnitude.Data[x + y * orientation.Stride] =
(float)
Math.Sqrt(imageX.Data[y * imageX.Stride + x] * imageX.Data[y * imageX.Stride + x] +
imageY.Data[y * imageY.Stride + x] * imageY.Data[y * imageY.Stride + x]);
}
}
return(new Tuple <ZagImage <int>, ZagImage <float> >(orientation, magnitude));
}
private ZagImage <byte> Convolution(ZagImage <byte> myImage, ConvMatrix Mask)
{
int s = Mask.Size / 2;
var cImage = myImage.Copy();
for (int y = s; y < myImage.Height - s; y++)
{
for (int x = s; x < myImage.Width - s; x++)
{
int r = 0, g = 0, b = 0;
for (int masky = 0; masky < Mask.Size; masky++)
{
for (int maskx = 0; maskx < Mask.Size; maskx++)
{
var maskValue = Mask.Matrix[maskx + masky * Mask.Size];
b += maskValue * myImage.Data[myImage.Depth * (x + maskx - s) + (y + masky - s) * myImage.Stride];
if (myImage.Depth == 3)
{
r += maskValue * myImage.Data[myImage.Depth * (x + maskx - s) + 2 + (y + masky - s) * myImage.Stride];
g += maskValue * myImage.Data[myImage.Depth * (x + maskx - s) + 1 + (y + masky - s) * myImage.Stride];
}
}
}
b = Math.Min(Math.Max((b / Mask.Factor) + Mask.Offset, 0), 255);
cImage.Data[myImage.Depth * x + y * myImage.Stride] = (byte)b;
if (myImage.Depth == 3)
{
r = Math.Min(Math.Max((r / Mask.Factor) + Mask.Offset, 0), 255);
g = Math.Min(Math.Max((g / Mask.Factor) + Mask.Offset, 0), 255);
cImage.Data[myImage.Depth * x + 1 + y * myImage.Stride] = (byte)g;
cImage.Data[myImage.Depth * x + 2 + y * myImage.Stride] = (byte)r;
}
}
}
return(cImage);
}
public ZagImage <byte> SmoothGaussian(ZagImage <byte> img)
{
var gaussMatrix = new ConvMatrix(new[] { 1, 2, 1, 2, 4, 2, 1, 2, 1 }, 3);
return(Convolution(img, gaussMatrix));
}
