/// <summary>
/// Fit ellipse through contour points in a least square sense.
/// </summary>
/// <param name="c">Contour</param>
/// <returns>Ellipse</returns>
private Emgu.CV.Structure.Ellipse FitByContour(Emgu.CV.Contour <System.Drawing.Point> c)
{
System.Drawing.PointF[] mypoints = Array.ConvertAll(
c.ToArray <System.Drawing.Point>(),
value => new System.Drawing.PointF(value.X, value.Y)
);
Emgu.CV.Structure.Ellipse e = Emgu.CV.PointCollection.EllipseLeastSquareFitting(mypoints);
Emgu.CV.Structure.MCvBox2D box = e.MCvBox2D;
// Modify ellipse to have width = a, height = b in the ellipse equation.
box.size.Height *= 0.5f;
box.size.Width *= 0.5f;
return(new Emgu.CV.Structure.Ellipse(box));
}
public void AddNeighbor(List <TessResult> tessOcrResultList, string inputpath)
{
string file_name = "";
List <PointF> points = new List <PointF>();
Boolean Continue = true;
int Counter = 0;
SVM SVMForClassification = new SVM(@"C:\Users\nhonarva\Documents\MachineLearning\sampleTraintData\");
for (int k = 0; k < tessOcrResultList.Count; k++) /// add neighbors of each bounding box
{
points = new List <PointF>();
string TotalPixelToSizeRatio = "";
file_name = tessOcrResultList[k].fileName;
Bitmap ResultOfTextDetection;
string[] split = file_name.Split('_');
if (split[6].Equals("0"))
{
ResultOfTextDetection = new Bitmap(inputpath + file_name + ".png");
}
else
{
string newFileName = "";
split[6] = "0";
for (int j = 0; j < split.Length; j++)
{
if (j < split.Length - 1)
{
newFileName += split[j] + "_";
}
else
{
newFileName += split[j];
}
}
ResultOfTextDetection = new Bitmap(inputpath + newFileName + ".png");
}
Bitmap OriginalBinaryImage = new Bitmap(inputpath + "BinaryOutput.png");
for (int i = 0; i < ResultOfTextDetection.Width; i++) //// scan image to find foreground pixels
{
for (int j = 0; j < ResultOfTextDetection.Height; j++)
{
Color color = ResultOfTextDetection.GetPixel(i, j);
if (!ResultOfTextDetection.GetPixel(i, j).Name.Equals("ffffffff"))
{
PointF point = new PointF(i, j);
points.Add(point);
}
}
}
String[] splitTokens = file_name.Split('_');
int numberOfRecognizedLetters = tessOcrResultList[k].tess_word3.Length - 2; /// -2 because we have \n\n at the end of word
int x = Convert.ToInt32(tessOcrResultList[k].x);
int y = Convert.ToInt32(tessOcrResultList[k].y);
int width = Convert.ToInt32(tessOcrResultList[k].w);
int height = Convert.ToInt32(tessOcrResultList[k].h);
if (splitTokens[6].Equals("0")) /// bounding box with slope 0 (horizental)
{
int estimatedSizeOfOneLetter = width / numberOfRecognizedLetters;
double pixelToSizeRatio = (double)(points.Count) / (double)(width * height);
TotalPixelToSizeRatio = Convert.ToString(Math.Round((decimal)pixelToSizeRatio, 2));
Point left = new Point(); /// left coordinate for box at left hand side of blob
int leftPixelCount = 0;
double leftPixelToSizeRatio = 0;
left.X = x - estimatedSizeOfOneLetter;
left.Y = y;
Point right = new Point(); //// right coordinate for box at right hand side of blob
int rightPixelCount = 0;
double rightPixelToSizeRatio = 0;
right.X = x + width;
right.Y = y;
Continue = true;
if (char.IsUpper(tessOcrResultList[k].tess_word3.ToCharArray()[0]))
{
Continue = false;
}
if (left.X < 0)
{
Continue = false;
}
// continue;
int AddedAreaWidth = estimatedSizeOfOneLetter;
Counter = 0;
while (Continue)
{
if ((left.X - estimatedSizeOfOneLetter) < 0)
{
AddedAreaWidth = left.X;
}
int[,] AddedAreaPoints = new int[AddedAreaWidth, height];
for (int i = left.X; i < left.X + AddedAreaWidth; i++)
{
for (int j = left.Y; j < left.Y + height; j++)
{
if ((!OriginalBinaryImage.GetPixel(i, j).Name.Equals("ffffffff")))
{
leftPixelCount++;
AddedAreaPoints[i - left.X, j - left.Y] = 1;
}
else
{
AddedAreaPoints[i - left.X, j - left.Y] = 0;
}
}
}
Counter++;
leftPixelToSizeRatio = (double)(leftPixelCount) / (double)(AddedAreaWidth * height);
QGISJsonNeighbothood.AddFeature(left.X, left.Y, left.X + AddedAreaWidth, left.Y, left.X + AddedAreaWidth, left.Y + tessOcrResultList[k].h, left.X, left.Y + tessOcrResultList[k].h, tessOcrResultList[k].id + "-" + Convert.ToString(Counter), Convert.ToString(Math.Round((decimal)leftPixelToSizeRatio, 2)), -1);
left.X -= estimatedSizeOfOneLetter;
leftPixelCount = 0;
if (left.X < 0)
{
Continue = false;
}
if (leftPixelToSizeRatio < 0.2)
{
Continue = false;
}
if (leftPixelToSizeRatio > 0.2)
{
SVMForClassification.Classification(AddedAreaPoints, tessOcrResultList[k].id + "-" + Convert.ToString(Counter));
}
}
Continue = true;
if (right.X > OriginalBinaryImage.Width)
{
Continue = false;
}
AddedAreaWidth = estimatedSizeOfOneLetter;
Counter = 0;
while (Continue)
{
if ((right.X + estimatedSizeOfOneLetter) > OriginalBinaryImage.Width)
{
AddedAreaWidth = OriginalBinaryImage.Width - right.X;
}
int[,] AddedAreaPoints = new int[AddedAreaWidth, height];
for (int i = right.X; i < right.X + AddedAreaWidth; i++)
{
for (int j = right.Y; j < right.Y + height; j++)
{
if (!OriginalBinaryImage.GetPixel(i, j).Name.Equals("ffffffff"))
{
rightPixelCount++;
AddedAreaPoints[i - right.X, j - right.Y] = 1;
}
else
{
AddedAreaPoints[i - right.X, j - right.Y] = 0;
}
}
}
Counter++;
rightPixelToSizeRatio = (double)(rightPixelCount) / (double)(AddedAreaWidth * height);
QGISJsonNeighbothood.AddFeature(right.X, right.Y, right.X + AddedAreaWidth, right.Y, right.X + AddedAreaWidth, right.Y + tessOcrResultList[k].h, right.X, right.Y + tessOcrResultList[k].h, tessOcrResultList[k].id + "-" + Convert.ToString(Counter), Convert.ToString(Math.Round((decimal)rightPixelToSizeRatio, 2)), -1);
right.X += estimatedSizeOfOneLetter;
rightPixelCount = 0;
if (right.X > OriginalBinaryImage.Width)
{
Continue = false;
}
if (rightPixelToSizeRatio < 0.2)
{
Continue = false;
}
if (rightPixelToSizeRatio > 0.2)
{
SVMForClassification.Classification(AddedAreaPoints, tessOcrResultList[k].id + "-" + Convert.ToString(Counter));
}
}
// }
QGISJsonNeighbothood.AddFeature(x, y, x + width, y, x + width, y + height, x, y + height, tessOcrResultList[k].id, TotalPixelToSizeRatio, -1);
}
else
{
PointF[] pts;
pts = points.ToArray();
Emgu.CV.Structure.MCvBox2D box = PointCollection.MinAreaRect(pts);
Rectangle boundingbox = PointCollection.BoundingRectangle(pts);
Emgu.CV.Structure.MCvBox2D boxinoriginalImage = box;
boxinoriginalImage.center.X += x - boundingbox.X; /// 63=originalboundingbox.x(1069)(I used result of intermediate image)-boundingbox.x(21)
boxinoriginalImage.center.Y += y - boundingbox.Y;
PointF[] vertices = boxinoriginalImage.GetVertices();
double pixelToSizeRatio = (double)(points.Count) / (double)(box.size.Width * box.size.Height);
TotalPixelToSizeRatio = Convert.ToString(Math.Round((decimal)pixelToSizeRatio, 2));
QGISJsonNeighbothood.AddFeature(vertices[0].X, vertices[0].Y, vertices[1].X, vertices[1].Y, vertices[2].X, vertices[2].Y, vertices[3].X, vertices[3].Y, tessOcrResultList[k].id, TotalPixelToSizeRatio, -1);
for (int m = 1; m < 4; m++)
{
Emgu.CV.Structure.MCvBox2D addedArea = box;
if (box.size.Height > box.size.Width) ///Non horizental bounding boxes that slope of bounding box is less than zero
{
double angleInRadius = ((90 + box.angle) * Math.PI) / 180;
double addToX = (((box.size.Height) * ((m - 1) * 2 + numberOfRecognizedLetters + 1)) / (numberOfRecognizedLetters * 2)) * Math.Cos(angleInRadius); // I have multiplied height to 3/4 because we have four letter in recognized image and I want to find the center of added part for boundingbox
double addToY = (((box.size.Height) * ((m - 1) * 2 + numberOfRecognizedLetters + 1)) / (numberOfRecognizedLetters * 2)) * Math.Sin(angleInRadius);
addedArea.center.X += (x - boundingbox.X) + (float)addToX;
addedArea.center.Y += (y - boundingbox.Y) + (float)addToY;
addedArea.size.Height = (float)((box.size.Height) / numberOfRecognizedLetters);
}
else ///Non horizental bounding boxes that slope of bounding box is greater than zero
{
double angleInRadius = (((-1) * box.angle) * Math.PI) / 180;
double addToX = (((box.size.Width) * ((m - 1) * 2 + numberOfRecognizedLetters + 1)) / (numberOfRecognizedLetters * 2)) * Math.Cos(angleInRadius); // I have multiplied height to 3/4 because we have four letter in recognized image and I want to find the center of added part for boundingbox
double addToY = (((box.size.Width) * ((m - 1) * 2 + numberOfRecognizedLetters + 1)) / (numberOfRecognizedLetters * 2)) * Math.Sin(angleInRadius);
addedArea.center.X += (x - boundingbox.X) - (float)addToX;
addedArea.center.Y += (y - boundingbox.Y) + (float)addToY;
addedArea.size.Width = (float)((box.size.Width) / numberOfRecognizedLetters);
}
PointF[] AddedAreaVertices = addedArea.GetVertices();
PointF ExploreAddedArea_MinPoint = AddedAreaVertices[0];
PointF ExploreAddedArea_MaxPoint = AddedAreaVertices[0];
for (int i = 1; i < AddedAreaVertices.Length; i++)
{
if (AddedAreaVertices[i].X < ExploreAddedArea_MinPoint.X)
{
ExploreAddedArea_MinPoint.X = AddedAreaVertices[i].X;
}
if (AddedAreaVertices[i].Y < ExploreAddedArea_MinPoint.Y)
{
ExploreAddedArea_MinPoint.Y = AddedAreaVertices[i].Y;
}
if (AddedAreaVertices[i].X > ExploreAddedArea_MaxPoint.X)
{
ExploreAddedArea_MaxPoint.X = AddedAreaVertices[i].X;
}
if (AddedAreaVertices[i].Y > ExploreAddedArea_MaxPoint.Y)
{
ExploreAddedArea_MaxPoint.Y = AddedAreaVertices[i].Y;
}
}
int AddedAreaPixelNumber = 0;
double[] slope = new double[4];
double[] b = new double[4];
for (int i = 0; i < AddedAreaVertices.Length; i++) //// find four lines that create min area rectangle bounding box
{
if (i < AddedAreaVertices.Length - 1)
{
slope[i] = (AddedAreaVertices[i].Y - AddedAreaVertices[i + 1].Y) / (AddedAreaVertices[i].X - AddedAreaVertices[i + 1].X);
}
else
{
slope[i] = (AddedAreaVertices[i].Y - AddedAreaVertices[0].Y) / (AddedAreaVertices[i].X - AddedAreaVertices[0].X);
}
b[i] = -(AddedAreaVertices[i].Y - slope[i] * AddedAreaVertices[i].X);
}
for (int i = Convert.ToInt32(ExploreAddedArea_MinPoint.X); i < Convert.ToInt32(ExploreAddedArea_MaxPoint.X) + 1; i++)
{
for (int j = Convert.ToInt32(ExploreAddedArea_MinPoint.Y); j < Convert.ToInt32(ExploreAddedArea_MaxPoint.Y) + 1; j++)
{
if ((i < OriginalBinaryImage.Width) && (j < OriginalBinaryImage.Width) && (i > 0) && (j > 0))
{
Color color = OriginalBinaryImage.GetPixel(i, j);
double[] result = new double[4];
for (int l = 0; l < result.Length; l++)
{
result[l] = j - slope[l] * i + b[l];
}
if (result[0] < 0 && result[1] > 0 && result[2] > 0 && result[3] < 0) /// explore if pixel is in bounding box of Min Area Rectangle
{
// sourceImage1.SetPixel(i, j, Color.Green);
if (!OriginalBinaryImage.GetPixel(i, j).Name.Equals("ffffffff"))
{
AddedAreaPixelNumber++;
}
}
}
else
{
break;
}
}
}
double addedAreaSize = addedArea.size.Height * addedArea.size.Width;
double AddedAreaPixelToSizeRatio = (double)(AddedAreaPixelNumber) / (addedArea.size.Height * addedArea.size.Width);
QGISJsonNeighbothood.AddFeature(AddedAreaVertices[0].X, AddedAreaVertices[0].Y, AddedAreaVertices[1].X, AddedAreaVertices[1].Y, AddedAreaVertices[2].X, AddedAreaVertices[2].Y, AddedAreaVertices[3].X, AddedAreaVertices[3].Y, tessOcrResultList[k].id, Convert.ToString(Math.Round((decimal)AddedAreaPixelToSizeRatio, 2)), -1);
}
for (int m = 1; m < 4; m++)
{
Emgu.CV.Structure.MCvBox2D addedArea = box;
if (box.size.Height > box.size.Width) ///Non horizental bounding boxes that slope of bounding box is less than zero
{
double angleInRadius = ((90 + box.angle) * Math.PI) / 180;
double addToX = (((box.size.Height) * ((m - 1) * 2 + numberOfRecognizedLetters + 1)) / (numberOfRecognizedLetters * 2)) * Math.Cos(angleInRadius); // I have multiplied height to 3/4 because we have four letter in recognized image and I want to find the center of added part for boundingbox
double addToY = (((box.size.Height) * ((m - 1) * 2 + numberOfRecognizedLetters + 1)) / (numberOfRecognizedLetters * 2)) * Math.Sin(angleInRadius);
addedArea.center.X += (x - boundingbox.X) - (float)addToX;
addedArea.center.Y += (y - boundingbox.Y) - (float)addToY;
addedArea.size.Height = (float)((box.size.Height) / numberOfRecognizedLetters);
}
else ///Non horizental bounding boxes that slope of bounding box is greater than zero
{
double angleInRadius = (((-1) * box.angle) * Math.PI) / 180;
double addToX = (((box.size.Width) * ((m - 1) * 2 + numberOfRecognizedLetters + 1)) / (numberOfRecognizedLetters * 2)) * Math.Cos(angleInRadius); // I have multiplied height to 3/4 because we have four letter in recognized image and I want to find the center of added part for boundingbox
double addToY = (((box.size.Width) * ((m - 1) * 2 + numberOfRecognizedLetters + 1)) / (numberOfRecognizedLetters * 2)) * Math.Sin(angleInRadius);
addedArea.center.X += (x - boundingbox.X) + (float)addToX;
addedArea.center.Y += (y - boundingbox.Y) - (float)addToY;
addedArea.size.Width = (float)((box.size.Width) / numberOfRecognizedLetters);
}
PointF[] AddedAreaVertices = addedArea.GetVertices();
PointF ExploreAddedArea_MinPoint = AddedAreaVertices[0];
PointF ExploreAddedArea_MaxPoint = AddedAreaVertices[0];
for (int i = 1; i < AddedAreaVertices.Length; i++)
{
if (AddedAreaVertices[i].X < ExploreAddedArea_MinPoint.X)
{
ExploreAddedArea_MinPoint.X = AddedAreaVertices[i].X;
}
if (AddedAreaVertices[i].Y < ExploreAddedArea_MinPoint.Y)
{
ExploreAddedArea_MinPoint.Y = AddedAreaVertices[i].Y;
}
if (AddedAreaVertices[i].X > ExploreAddedArea_MaxPoint.X)
{
ExploreAddedArea_MaxPoint.X = AddedAreaVertices[i].X;
}
if (AddedAreaVertices[i].Y > ExploreAddedArea_MaxPoint.Y)
{
ExploreAddedArea_MaxPoint.Y = AddedAreaVertices[i].Y;
}
}
int AddedAreaPixelNumber = 0;
double[] slope = new double[4];
double[] b = new double[4];
for (int i = 0; i < AddedAreaVertices.Length; i++) //// find four lines that create min area rectangle bounding box
{
if (i < AddedAreaVertices.Length - 1)
{
slope[i] = (AddedAreaVertices[i].Y - AddedAreaVertices[i + 1].Y) / (AddedAreaVertices[i].X - AddedAreaVertices[i + 1].X);
}
else
{
slope[i] = (AddedAreaVertices[i].Y - AddedAreaVertices[0].Y) / (AddedAreaVertices[i].X - AddedAreaVertices[0].X);
}
b[i] = -(AddedAreaVertices[i].Y - slope[i] * AddedAreaVertices[i].X);
}
for (int i = Convert.ToInt32(ExploreAddedArea_MinPoint.X); i < Convert.ToInt32(ExploreAddedArea_MaxPoint.X) + 1; i++)
{
for (int j = Convert.ToInt32(ExploreAddedArea_MinPoint.Y); j < Convert.ToInt32(ExploreAddedArea_MaxPoint.Y) + 1; j++)
{
if ((i < OriginalBinaryImage.Width) && (j < OriginalBinaryImage.Width) && (i > 0) && (j > 0))
{
Color color = OriginalBinaryImage.GetPixel(i, j);
double[] result = new double[4];
for (int l = 0; l < result.Length; l++)
{
result[l] = j - slope[l] * i + b[l];
}
if (result[0] < 0 && result[1] > 0 && result[2] > 0 && result[3] < 0) /// explore if pixel is in bounding box of Min Area Rectangle
{
// sourceImage1.SetPixel(i, j, Color.Green);
if (!OriginalBinaryImage.GetPixel(i, j).Name.Equals("ffffffff"))
{
AddedAreaPixelNumber++;
}
}
}
else
{
break;
}
}
}
double addedAreaSize = addedArea.size.Height * addedArea.size.Width;
double AddedAreaPixelToSizeRatio = (double)(AddedAreaPixelNumber) / (addedArea.size.Height * addedArea.size.Width);
QGISJsonNeighbothood.AddFeature(AddedAreaVertices[0].X, AddedAreaVertices[0].Y, AddedAreaVertices[1].X, AddedAreaVertices[1].Y, AddedAreaVertices[2].X, AddedAreaVertices[2].Y, AddedAreaVertices[3].X, AddedAreaVertices[3].Y, tessOcrResultList[k].id, Convert.ToString(Math.Round((decimal)AddedAreaPixelToSizeRatio, 2)), -1);
}
}
}
QGISJsonNeighbothood.WriteGeojsonFileByPixels();
}
