CODEWITHSUNDEEP
c#emgucv
user3389475
user3389475

Reputation: 174

How do I find images inside an image?

How do I find images inside an image?

Right now I'm using EmguCV and I have cooked up the following code based on this Tutorial: https://www.emgu.com/wiki/index.php/Shape_(Triangle,_Rectangle,_Circle,_Line)_Detection_in_CSharp

        using (UMat gray = new UMat())
        using (UMat cannyEdges = new UMat())
        {
            CvInvoke.CvtColor(img, gray, ColorConversion.Bgr2Gray);
            CvInvoke.GaussianBlur(gray, gray, new Size(3, 3), 2);
            CvInvoke.Canny(gray, cannyEdges, 0, 10, 3);

            LineSegment2D[] lines = CvInvoke.HoughLinesP(
                cannyEdges,
                1, //Distance resolution in pixel-related units
                Math.PI / 2, //Angle resolution measured in radians.
                0, //threshold
                40, //min Line width
                1); //gap between lines


            foreach (LineSegment2D line in lines)
            {
                CvInvoke.Line(img, line.P1, line.P2, new Bgr(Color.Red).MCvScalar, 1);
            }
        }

Step 1 enter image description here Step 2 enter image description here Step 3 enter image description here

This is my best result so far, but it's not perfect since i still need some edges to be able to create all bounding boxes around the images.

Step 4 enter image description here

What I want, is to find all edges of every single images in the image so I can make perfect bounding boxes like so. The red areas are what i want to find

I'm new to EmguCV/OpenCV, but i still think my best bet is on this library for solving this problem. I just need to find the right tools and use them right, and that's what I hope someone here can help me with :)

Upvotes: 0

Views: 991

Answers (1)

Oliver
Oliver

Reputation: 696

Here's what you can do:

  • Load the image
  • Invert - img.Not()
  • Convert to Grayscale - img.Convert<Gray,byte>()
  • Perform binary thresholding - img.Convert<Gray, byte>().ThresholdBinary(new Gray(54), new Gray(255))

It will result to the following image, optimize the threshold value to get better results: enter image description here

  • Get the convex hull and bounding box. Perform filtering so you get the objects that fits your criteria. You can use the contour area, perimeter, etc.

    var Contours = new List<Contour<Point>>();
for (Contour<Point> contours = _gray.FindContours(
   HAIN_APPROX_METHOD.CV_CHAIN_APPROX_SIMPLE, RETR_TYPE.CV_RETR_EXTERNAL); 
   contours != null; contours = contours.HNext)
{
    Seq<Point> pts = contours.GetConvexHull(ORIENTATION.CV_CLOCKWISE);
    double diff = Math.Round(Math.Abs(pts.Area - contours.Area) / 
      pts.Area, 2);

    //additional constraint
    double q = contours.Area / contours.Perimeter;

    //bounding box of the counter
    Rectangle rect = contours.BoundingRectangle;

    //customize the value to suit your need
    if (contours.BoundingRectangle.Height > 5 && 
     contours.BoundingRectangle.Width > 5)
    {
        Contours.Add(contours);
    }
}

  • Do what you want with Contours, (e.g. fill and use as mask to extract each image, draw bounding box, etc.)

Upvotes: 1

Related Questions