Question

I’m trying to find a way to do an approximated segmentation on graveyard images (in context of CBIR in cultural scienes – but that’s not the topic). So far I’m using this strategy:

• Blurr image twice (experimental findings)
• Apply Canny-Edge-Detector

• Finding the contours

  int main(int argc, const char* argv[]) {
  
  cout << "Starting " << endl;
  Mat sourceImage;
  sourceImage = imread("singlesegmentation/DSCN5204.JPG",
          CV_LOAD_IMAGE_COLOR);
  
  if (!sourceImage.data) {
      cout << "No Image found." << endl;
      return -1;
  }
  
  cv::Mat blurred = imagePro::blurrNtimes(2, sourceImage);
  cv::Mat target = edged::applyCanny(blurred);
  cout << "Canny applied " << endl;
  
  vector<vector<Point> > contours;
  vector<Vec4i> hierarchy;
  cv::Point offset;
  offset.x = sourceImage.rows / 2;
  offset.y = sourceImage.cols / 2;
  cv::findContours(target, contours, hierarchy, CV_RETR_TREE ,
          CV_CHAIN_APPROX_SIMPLE, offset);
  cout << "Contours applied " << endl;
  
  int idx = 0;
  for (; idx >= 0; idx = hierarchy[idx][0]) {
      Scalar color(rand() & 255, rand() & 255, rand() & 255);
      drawContours(target, contours, idx, color, CV_FILLED, 8, hierarchy);
  }
  cout << "Lines applied " << endl;
  
  cv::namedWindow("Contour", CV_WINDOW_NORMAL);
  cv::imshow("Contour", target);
  cv::waitKey(0);
  
  return 0;
  

  }

The namespaces “imagePro” and “edged” contain simple code of opencv to blur a image and further process it. The code works. Here is an example-picture:

But now I have no idea to segment the image. I want to go from the inside to the outside of the rectangle-stone and when I find a line I want to remember the coordinates and then cut the content. Thank you if you have an idea or hint!

Answer 1

You could try use a Hough-Transformation (cv::HoughLinesP) see the tutorial examples http://docs.opencv.org/modules/imgproc/doc/feature_detection.html

In order to find the coordinates of the stone wihtin the picture you would need to calculate the intersections of the lines found by the hough-transformation. I used Gaussian-Blur followed by Laplace-Transformation (instead of canny-edge) for a similar use case.