private static IEnumerable <TextLine> GetLines(List <Word> words, double maxDist, AngleBounds withinLine)
{
TextDirection textDirection = words[0].TextDirection;
var groupedIndexes = ClusteringAlgorithms.ClusterNearestNeighbours(words, Distances.Euclidean,
(pivot, candidate) => maxDist,
pivot => pivot.BoundingBox.BottomRight, candidate => candidate.BoundingBox.BottomLeft,
pivot => true,
(pivot, candidate) =>
{
// Compare bottom right with bottom left for angle
var withinLineAngle = Distances.Angle(pivot.BoundingBox.BottomRight, candidate.BoundingBox.BottomLeft);
return(withinLineAngle >= withinLine.Lower && withinLineAngle <= withinLine.Upper);
}).ToList();
Func <IEnumerable <Word>, IReadOnlyList <Word> > orderFunc = l => l.OrderBy(x => x.BoundingBox.Left).ToList();
if (textDirection == TextDirection.Rotate180)
{
orderFunc = l => l.OrderByDescending(x => x.BoundingBox.Right).ToList();
}
else if (textDirection == TextDirection.Rotate90)
{
orderFunc = l => l.OrderByDescending(x => x.BoundingBox.Top).ToList();
}
else if (textDirection == TextDirection.Rotate270)
{
orderFunc = l => l.OrderBy(x => x.BoundingBox.Bottom).ToList();
}
for (var a = 0; a < groupedIndexes.Count; a++)
{
yield return(new TextLine(orderFunc(groupedIndexes[a].Select(i => words[i]))));
}
}
/// <summary>
/// Get information on the nearest point, filtered for angle.
/// </summary>
/// <param name="words"></param>
/// <param name="pivot"></param>
/// <param name="funcPivotDist"></param>
/// <param name="funcPivotAngle"></param>
/// <param name="funcPointsDist"></param>
/// <param name="funcPointsAngle"></param>
/// <param name="angleStart"></param>
/// <param name="angleEnd"></param>
/// <param name="finalDistMEasure"></param>
/// <returns></returns>
private double[] GetNearestPointData(Word[] words, Word pivot, Func <PdfRectangle,
PdfPoint> funcPivotDist, Func <PdfRectangle, PdfPoint> funcPivotAngle,
Func <PdfRectangle, PdfPoint> funcPointsDist, Func <PdfRectangle, PdfPoint> funcPointsAngle,
double angleStart, double angleEnd,
Func <PdfPoint, PdfPoint, double> finalDistMEasure)
{
var pointR = funcPivotDist(pivot.BoundingBox);
// Filter by angle
var filtered = words.Where(w =>
{
var angleWL = Distances.Angle(funcPivotAngle(pivot.BoundingBox), funcPointsAngle(w.BoundingBox));
return(angleWL >= angleStart && angleWL <= angleEnd);
}).ToList();
filtered.Remove(pivot); // remove itself
if (filtered.Count > 0)
{
int index = pointR.FindIndexNearest(
filtered.Select(w => funcPointsDist(w.BoundingBox)).ToList(),
Distances.Euclidean, out double distWL);
if (index >= 0)
{
var matchWL = filtered[index];
return(new double[]
{
(double)pivot.Letters.Select(l => l.FontSize).Mode(),
finalDistMEasure(pointR, funcPointsDist(matchWL.BoundingBox))
});
}
}
return(null);
}
private PdfRectangle GetBoundingBoxOther(IReadOnlyList <TextLine> lines)
{
var points = lines.SelectMany(l => new[]
{
l.BoundingBox.BottomLeft,
l.BoundingBox.BottomRight,
l.BoundingBox.TopLeft,
l.BoundingBox.TopRight
});
// Candidates bounding boxes
var obb = Geometry.GeometryExtensions.MinimumAreaRectangle(points);
var obb1 = new PdfRectangle(obb.BottomLeft, obb.TopLeft, obb.BottomRight, obb.TopRight);
var obb2 = new PdfRectangle(obb.BottomRight, obb.BottomLeft, obb.TopRight, obb.TopLeft);
var obb3 = new PdfRectangle(obb.TopRight, obb.BottomRight, obb.TopLeft, obb.BottomLeft);
// Find the orientation of the OBB, using the baseline angle
// Assumes line order is correct
var lastLine = lines[lines.Count - 1];
var baseLineAngle = Distances.BoundAngle180(Distances.Angle(lastLine.BoundingBox.BottomLeft, lastLine.BoundingBox.BottomRight));
double deltaAngle = Math.Abs(Distances.BoundAngle180(obb.Rotation - baseLineAngle));
double deltaAngle1 = Math.Abs(Distances.BoundAngle180(obb1.Rotation - baseLineAngle));
if (deltaAngle1 < deltaAngle)
{
deltaAngle = deltaAngle1;
obb = obb1;
}
double deltaAngle2 = Math.Abs(Distances.BoundAngle180(obb2.Rotation - baseLineAngle));
if (deltaAngle2 < deltaAngle)
{
deltaAngle = deltaAngle2;
obb = obb2;
}
double deltaAngle3 = Math.Abs(Distances.BoundAngle180(obb3.Rotation - baseLineAngle));
if (deltaAngle3 < deltaAngle)
{
obb = obb3;
}
return(obb);
}
/// <summary>
/// Get information on the nearest point, filtered for angle.
/// </summary>
private double?GetNearestPointDistance(List <Word> words, Word pivot, Func <PdfRectangle,
PdfPoint> funcPivotDist, Func <PdfRectangle, PdfPoint> funcPivotAngle,
Func <PdfRectangle, PdfPoint> funcPointsDist, Func <PdfRectangle, PdfPoint> funcPointsAngle,
AngleBounds angleBounds,
Func <PdfPoint, PdfPoint, double> finalDistanceMeasure)
{
var pointR = funcPivotDist(pivot.BoundingBox);
var pivotPoint = funcPivotAngle(pivot.BoundingBox);
var wordsWithinAngleBoundDistancePoints = new List <PdfPoint>();
// Filter to words within the angle range.
foreach (var word in words)
{
// Ignore the pivot word.
if (ReferenceEquals(word, pivot))
{
continue;
}
var angle = Distances.Angle(pivotPoint, funcPointsAngle(word.BoundingBox));
if (angleBounds.Contains(angle))
{
wordsWithinAngleBoundDistancePoints.Add(funcPointsDist(word.BoundingBox));
}
}
if (wordsWithinAngleBoundDistancePoints.Count == 0)
{
return(null);
}
var closestWordIndex = pointR.FindIndexNearest(wordsWithinAngleBoundDistancePoints, p => p,
p => p, Distances.Euclidean, out _);
if (closestWordIndex < 0 || closestWordIndex >= wordsWithinAngleBoundDistancePoints.Count)
{
return(null);
}
return(finalDistanceMeasure(pointR, wordsWithinAngleBoundDistancePoints[closestWordIndex]));
}
/// <summary>
/// Build lines via transitive closure.
/// </summary>
private static IEnumerable <TextLine> GetLines(List <Word> words, double maxDist, AngleBounds withinLine)
{
/***************************************************************************************************
* /!\ WARNING: Given how FindIndexNearest() works, if 'maxDist' > 'word Width', the algo might not
* work as the FindIndexNearest() function might pair the pivot with itself (the pivot's right point
* (distance = width) is closer than other words' left point).
* -> Solution would be to find more than one nearest neighbours. Use KDTree?
***************************************************************************************************/
TextDirection textDirection = words[0].TextDirection;
var groupedIndexes = ClusteringAlgorithms.SimpleTransitiveClosure(words, Distances.Euclidean,
(pivot, candidate) => maxDist,
pivot => pivot.BoundingBox.BottomRight, candidate => candidate.BoundingBox.BottomLeft,
pivot => true,
(pivot, candidate) =>
{
// Compare bottom right with bottom left for angle
var withinLineAngle = Distances.Angle(pivot.BoundingBox.BottomRight, candidate.BoundingBox.BottomLeft);
return(withinLineAngle >= withinLine.Lower && withinLineAngle <= withinLine.Upper);
}).ToList();
Func <IEnumerable <Word>, IReadOnlyList <Word> > orderFunc = l => l.OrderBy(x => x.BoundingBox.Left).ToList();
if (textDirection == TextDirection.Rotate180)
{
orderFunc = l => l.OrderByDescending(x => x.BoundingBox.Right).ToList();
}
else if (textDirection == TextDirection.Rotate90)
{
orderFunc = l => l.OrderByDescending(x => x.BoundingBox.Top).ToList();
}
else if (textDirection == TextDirection.Rotate270)
{
orderFunc = l => l.OrderBy(x => x.BoundingBox.Bottom).ToList();
}
for (var a = 0; a < groupedIndexes.Count; a++)
{
yield return(new TextLine(orderFunc(groupedIndexes[a].Select(i => words[i]))));
}
}
private static PdfRectangle GetBoundingBoxOther(IReadOnlyList <Word> words)
{
var baseLinePoints = words.SelectMany(r => new[]
{
r.BoundingBox.BottomLeft,
r.BoundingBox.BottomRight,
}).ToList();
// Fitting a line through the base lines points
// to find the orientation (slope)
double x0 = baseLinePoints.Average(p => p.X);
double y0 = baseLinePoints.Average(p => p.Y);
double sumProduct = 0;
double sumDiffSquaredX = 0;
for (int i = 0; i < baseLinePoints.Count; i++)
{
var point = baseLinePoints[i];
var x_diff = point.X - x0;
var y_diff = point.Y - y0;
sumProduct += x_diff * y_diff;
sumDiffSquaredX += x_diff * x_diff;
}
double cos = 0;
double sin = 1;
if (sumDiffSquaredX > 1e-3)
{
// not a vertical line
double angleRad = Math.Atan(sumProduct / sumDiffSquaredX); // -π/2 ≤ θ ≤ π/2
cos = Math.Cos(angleRad);
sin = Math.Sin(angleRad);
}
// Rotate the points to build the axis-aligned bounding box (AABB)
var inverseRotation = new TransformationMatrix(
cos, -sin, 0,
sin, cos, 0,
0, 0, 1);
var transformedPoints = words.SelectMany(r => new[]
{
r.BoundingBox.BottomLeft,
r.BoundingBox.BottomRight,
r.BoundingBox.TopLeft,
r.BoundingBox.TopRight
}).Distinct().Select(p => inverseRotation.Transform(p));
var aabb = new PdfRectangle(transformedPoints.Min(p => p.X),
transformedPoints.Min(p => p.Y),
transformedPoints.Max(p => p.X),
transformedPoints.Max(p => p.Y));
// Rotate back the AABB to obtain to oriented bounding box (OBB)
var rotateBack = new TransformationMatrix(
cos, sin, 0,
-sin, cos, 0,
0, 0, 1);
// Candidates bounding boxes
var obb = rotateBack.Transform(aabb);
var obb1 = new PdfRectangle(obb.BottomLeft, obb.TopLeft, obb.BottomRight, obb.TopRight);
var obb2 = new PdfRectangle(obb.BottomRight, obb.BottomLeft, obb.TopRight, obb.TopLeft);
var obb3 = new PdfRectangle(obb.TopRight, obb.BottomRight, obb.TopLeft, obb.BottomLeft);
// Find the orientation of the OBB, using the baseline angle
// Assumes word order is correct
var firstWord = words[0];
var lastWord = words[words.Count - 1];
var baseLineAngle = Distances.Angle(firstWord.BoundingBox.BottomLeft, lastWord.BoundingBox.BottomRight);
double deltaAngle = Math.Abs(Distances.BoundAngle180(obb.Rotation - baseLineAngle));
double deltaAngle1 = Math.Abs(Distances.BoundAngle180(obb1.Rotation - baseLineAngle));
if (deltaAngle1 < deltaAngle)
{
deltaAngle = deltaAngle1;
obb = obb1;
}
double deltaAngle2 = Math.Abs(Distances.BoundAngle180(obb2.Rotation - baseLineAngle));
if (deltaAngle2 < deltaAngle)
{
deltaAngle = deltaAngle2;
obb = obb2;
}
double deltaAngle3 = Math.Abs(Distances.BoundAngle180(obb3.Rotation - baseLineAngle));
if (deltaAngle3 < deltaAngle)
{
obb = obb3;
}
return(obb);
}