Realisierung der Suche nach Briefmarken auf OpenCV ohne Neuronen, Registrierung und SMS

import sys
from collections import namedtuple

Metrics = namedtuple("metrics", ["tp", "fp", "fn"])

def bb_intersection_over_union(a, b):
   #           .
   # 0 -  X    ,
   # 1 -  Y    ,
   # 2 -  X    ,
   # 3 -  Y    .
   x_a = max(a[0], b[0])
   y_a = max(a[1], b[1])
   x_b = min(a[2], b[2])
   y_b = min(a[3], b[3])

   #  .
   inter_area = max(0, x_b - x_a + 1) * max(0, y_b - y_a + 1)

   #  .
   a_area = (a[2] - a[0] + 1) * (a[3] - a[1] + 1)
   b_area = (b[2] - b[0] + 1) * (b[3] - b[1] + 1)

   # Intersection over Union.
   iou = inter_area / float(a_area + b_area - inter_area)

   return iou

def same_stamp(a, b):
   """,         """
   iou = bb_intersection_over_union(a, b)
   print(f'iou: {iou}')
   return iou > 0.6

def compare_stamps(extracted, mapped):
   """
    
   :param extracted: ,  .
   :param mapped:  () .
   """
   tp = []
   fp = []
   fn = list(mapped)
   for stamp in extracted:
       for check in fn:
           if same_stamp(stamp, check):
               tp.append(check)
               fn.remove(check)
               break
       else:
           fp.append(stamp)
   return Metrics(len(tp), len(fp), len(fn))

def compare(file, sectors):
   """
       .
   :param file:   .
   :param sectors:  .
   """
   print(f'file: {file}')
   try:
       stamps = extract_stamps(file) #  ,     .
       metrics = compare_stamps(stamps, [ss for ss in sectors if 'stamp' in ss['tags']])
       return file, metrics
   except:
       print(sys.exc_info())
       return file, Metrics(0, 0, 0)

if __name__ == '__main__':
   file_metrics = {}

   # dataset - ,            () .
   for file, sectors in dataset.items():
       file_metrics[file] = compare(file, sectors)

   #  ,   F1-.
   total_metrics = Metrics(*(sum(x) for x in zip(*file_metrics.values())))
   precision = total_metrics.tp / (total_metrics.tp + total_metrics.fp) if total_metrics.tp > 0 else 0
   recall = total_metrics.tp / (total_metrics.tp + total_metrics.fn) if total_metrics.tp > 0 else 0
   f1 = 2 * precision * recall / (precision + recall) if (precision + recall) > 0 else 0

   print('precision\trecall\tf1')
   print('{:.4f}\t{:.4f}\t{:.4f}'.format(precision * 100, recall * 100, f1 * 100).replace('.', ','))
   print('tp\tfp\tfn')
   print('{}\t{}\t{}'.format(total_metrics.tp, total_metrics.fp, total_metrics.fn))

   print('tp\tfp\tfn')
   for file in dataset.keys():
       metric = file_metrics.get(file)
       print(f'{metric.tp}\t{metric.fp}\t{metric.fn}')

   print(f'precision: {precision}, recall: {recall}, f1: {f1}, {total_metrics}')

def colored_mask(img, threshold = -1):
   #     .
   denoised = cv2.medianBlur(img, 3)
   cv2.imwrite('denoised.bmp', denoised)

   #      .
   gray = cv2.cvtColor(denoised, cv2.COLOR_BGR2GRAY)
   cv2.imwrite('gray.bmp', gray)

   #    .
   adaptiveThreshold = threshold if threshold >= 0 else cv2.mean(img)[0]
   color = cv2.cvtColor(denoised, cv2.COLOR_BGR2HLS)
   mask = cv2.inRange(color, (0, int(adaptiveThreshold / 6), 60), (180, adaptiveThreshold, 255))

   #     .
   dst = cv2.bitwise_and(gray, gray, mask=mask)
   cv2.imwrite('colors_mask.bmp', dst)
   return dst

private void ColoredMask(Mat src, Mat dst, double threshold = -1)
{
 using (var gray = new Mat())
 using (var color = new Mat())
 using (var mask = new Mat())
 using (var denoised = new Mat())
 {
   //     .
   Cv2.MedianBlur(src, denoised, 3);
  denoised.Save("colors_denoised.bmp");

   //      .
   Cv2.CvtColor(denoised, gray, ColorConversionCodes.BGR2GRAY);
   gray.Save("colors_gray.bmp");

   //    .
   var adaptiveThreshold = threshold < 0 ? src.Mean()[0] : threshold;
   Cv2.CvtColor(denoised, color, ColorConversionCodes.BGR2HLS);
   Cv2.InRange(color, new Scalar(0, adaptiveThreshold / 6, 60), new Scalar(180, adaptiveThreshold, 255), mask);

   //     .
   Cv2.BitwiseAnd(gray, gray, dst, mask);
   dst.Save("colors_mask.bmp");
 }
}

# mask - ,   ;
# method:    ( OpenCV   Gradient);
# dp: ,      (1 —    , 2 —  /  ..     (..  ),   “”   - ,     .   ,   “”    );
# minDist:     (  %   );
# param1:   ,     (     2  );
# param2:     ;
# minRadius:   ;
# maxRadius:   ;
circles = cv2.HoughCircles(mask, cv2.HOUGH_GRADIENT, 1, 20, param1, param2, minRadius, maxRadius)

#   .
circles = np.uint16(np.around(circles))
cimg = cv2.cvtColor(img,cv2.COLOR_GRAY2BGR)
for i in circles[0,:]:
    cv2.circle(cimg,(i[0],i[1]),i[2],(165,25,165),2)

// mask - ,   ;
// method:    ( OpenCV   Gradient);
// dp: ,      (1 —    , 2 —  /  ..     (..  ),   “”   - ,     .   ,   “”    );
// minDist:     (  %   );
// param1:   ,     (     2  );
// param2:     ;
// minRadius:   ;
// maxRadius:   ;
var houghCircles = Cv2.HoughCircles(morphed, HoughMethods.Gradient, 1, max, cannyEdgeThreshold, houghThreshold, min, max);

//   .
foreach (var circle in houghCircles)
 Cv2.Circle(morphed, circle.Center, (int)circle.Radius, new Scalar(165, 25, 165), 2);

def equals(first, second, epsilon):
   diff = cv2.subtract(first, second)
   nonZero = cv2.countNonZero(diff)
   area = first.size * epsilon
   return nonZero <= area

for i in circles[0, :]:
   empty = np.zeros((256, 256, 1), dtype="uint8")
   cv2.circle(empty, (i[0], i[1]), i[2], (255, 255, 255), -1)
   crop = img * (empty.astype(img.dtype))

   cv2.imwrite('crop.bmp', crop)

   if not equals(crop, empty, threshold):
       result.append(i)

foreach (var circle in circles)
{
  var x = (int) circle.Center.X;
  var y = (int) circle.Center.Y;
  var radius = (int) Math.Floor(circle.Radius);

  using (var empty = Mat.Zeros(src.Rows, src.Cols, MatType.CV_8UC1))
  using (var mask = empty.ToMat())
  using (var crop = new Mat())
  {
    //   ()   .
    Cv2.Circle(mask, x, y, radius, Scalar.White, -1);
    src.CopyTo(crop, mask);

    crop.Save("crop.bmp");

    //    .
    if (!MatEquals(crop, empty, threshold))
    result.Add(circle);
  }
}

private bool MatEquals(Mat a, Mat b, double epsilon = 0.0)
{
  var notEquals = a.NotEquals(b);
  var nonZero = Cv2.CountNonZero(notEquals);
  var area = a.Rows * epsilon * a.Cols;
  return nonZero <= area;
}

segments = []
contours, hierarchy = cv2.findContours(img, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
for contour in contours:
   if contour.shape[0] < 5:
       continue

   ellipse = cv2.fitEllipse(contour)
   width = ellipse[1][0]
   height = ellipse[1][1]
   minor = min(width, height)
   major = max(width, height)

   if minor / 2 > minorMin and major / 2 < majorMax:
       r1 = math.fabs(1 - math.fabs(major - minor) / max(minor, major))
       cv2.ellipse(src, ellipse, (255, 0, 0), 3)

       if r1 > roundness:
           segments.append((ellipse[0], major / 2))
           cv2.ellipse(src, ellipse, (0, 255, 0), 3)
   else: cv2.ellipse(src, ellipse, (0, 0, 255), 1)

cv2.imwrite('test_res.bmp', src)

def distance(p1, p2):
   return math.sqrt(math.pow(p2[0] - p1[0], 2) + math.pow(p2[1] - p1[1], 2))

def isNested(inner, outer, epsilon):
   distance = distance(inner[0], outer[0])
   radius = outer[1] * epsilon
   return distance < radius and inner[1] < radius - distance

nested = []
for i in inner:
   for o in outer:
       if (isNested(i, o, 1.3)):
           if (distance(i[0], i[1]) < 30 and i[1] / o[1] > 0.75):
               nested.append(i)
           else: nested.append(o)

Cv2.FindContours(src, out var contours, hierarchy, RetrievalModes.External, ContourApproximationModes.ApproxSimple);
  foreach (var contour in contours)
  {
    //      5 .
    if (contour.Height < 5)
    continue;

    var ellipse = Cv2.FitEllipse(contour);
    var minor = Math.Min(ellipse.Size.Width, ellipse.Size.Height);
    var major = Math.Max(ellipse.Size.Width, ellipse.Size.Height);

    if (minor / 2 > minorSize && major / 2 < majorSize)
    {
    //     .   ,      .
    var r1 = Math.Abs(1 - Math.Abs(major - minor) / Math.Max(minor, major));
    if (r1 > roundness)
    {
        var circle = new CircleSegment(ellipse.Center, major / 2);
        segments.Add(circle);
    }
    }
    contour.Dispose();
  }
}

private bool IsCirclesNested(CircleSegment inner, CircleSegment outer, double epsilon)
{
  var distance = inner.Center.DistanceTo(outer.Center);
  var secondRadius = outer.Radius * epsilon;
  return distance < secondRadius && inner.Radius < secondRadius - distance;
}

var nested = new List<CircleSegment>();
foreach (var i in inner)
foreach (var o in outer)
{
  if (IsCirclesNested(i, o, 1.3))
  {
    if (i.Center.DistanceTo(o.Center) < 30 &&
        i.Radius / o.Radius > 0.75)
    nested.Add(i);
    else nested.Add(o);
  }
}

return outer.Union(inner).Except(nested).ToList();

maxRadius = math.sqrt(math.pow(src.shape[0], 2) + math.pow(src.shape[1], 2)) / 2
magnitude = src.shape[0] / math.log(maxRadius)
center = (src.shape[0] / 2, src.shape[1] / 2)
polar = cv2.logPolar(src, center, magnitude, cv2.INTER_AREA)

var maxRadius = Math.Sqrt(Math.Pow(stampImage.Width, 2) + Math.Pow(stampImage.Height, 2)) / 2;
var magnitude = stampImage.Width / Math.Log(maxRadius);
var center = new Point2f(stampImage.Width / 2, stampImage.Height / 2);
Cv2.LogPolar(stampImage, cartesianImage, center, magnitude, InterpolationFlags.Area);

#  .
sobel = cv2.Sobel(polar, cv2.CV_16S, 1, 0)
kernel = cv2.getStructuringElement(shape=cv2.MORPH_RECT, ksize=(1, 5))
img = cv2.convertScaleAbs(sobel)
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
img = cv2.threshold(img, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]
img = cv2.morphologyEx(img, cv2.MORPH_OPEN, kernel)
img = cv2.morphologyEx(img, cv2.MORPH_CLOSE, kernel)

#   .
def is_vertical(img_src, line):
   tolerance = 10
   coords = line[0]
   angle = math.atan2(coords[3] - coords[1], coords[2] - coords[0]) * 180.0 / math.pi
   edge = img_src.shape[0] * 0.66
   out_of_bounds = coords[0] < edge and coords[2] < edge
   return math.fabs(90 - math.fabs(angle)) <= tolerance and not out_of_bounds

lines = cv2.HoughLinesP(img, 1, math.pi / 180, 15, img.shape[0] / 5, 10)
vertical = [line for line in lines if is_vertical(img, line)]
correct_lines = len(vertical)

//  .
using (var sobel = new Mat())
using (var kernel = Cv2.GetStructuringElement(MorphShapes.Rect, new Size(1, 5)))
{
  Cv2.Sobel(img, sobel, MatType.CV_16S, 1, 0);
  Cv2.ConvertScaleAbs(sobel, img);
  Cv2.Threshold(img, img, 0, 255, ThresholdTypes.Binary | ThresholdTypes.Otsu);
  Cv2.MorphologyEx(img, img, MorphTypes.Open, kernel);
  Cv2.MorphologyEx(img, img, MorphTypes.Close, kernel);
}

//   .
bool AlmostVerticalLine(LineSegmentPoint line)
{
  const int tolerance = 10;
  var angle = Math.Atan2(line.P2.Y - line.P1.Y, line.P2.X - line.P1.X) * 180.0 / Math.PI;
  var edge = edges.Width * 0.66;
  var outOfBounds = line.P1.X < edge && line.P2.X < edge;
  return Math.Abs(90 - Math.Abs(angle)) <= tolerance && !outOfBounds;
}
var lines = Cv2.HoughLinesP(img, 1, Math.PI / 180, 15, img.Width / 5, 10);
var correctLinesCount = lines.Count(AlmostVerticalLine);