EmguCV C#：FindContours() 检测不同形状

Question

I have this image:

What I try to do is detecting the contours of it. So with looking to the documentation and some code on the web I made this:

Image<Gray, byte> image = receivedImage.Convert<Gray, byte>().ThresholdBinary(new Gray(80), new Gray(255));
        Emgu.CV.Util.VectorOfVectorOfPoint contours = new Emgu.CV.Util.VectorOfVectorOfPoint();
        Mat hier = new Mat();

        CvInvoke.FindContours(image, contours, hier, Emgu.CV.CvEnum.RetrType.External, Emgu.CV.CvEnum.ChainApproxMethod.ChainApproxSimple);

        CvInvoke.DrawContours(receivedImage, contours, 0, new MCvScalar(255, 0, 0), 2);

Then it detects this contour in blue:

Now I would like to detect both rectangles in differents contours. So the result would be this:

(made with paint) So now I would like to detect the two rectangles separetely, (the blue and red rectangles would be two different contours). But I have no idea about how to do that !

Thanks in advance for your help; ;)

Answer 1

The problem comes from the process of ThresholdBinary . As I assume you understand, this method will return a binary image, whereby all pixels above or equal to the threshold parameter, will be pulled up to maxValue parameter, and all those below, pulled down to 0. The produced image will consist therefore of only two values (binary), 0 or maxValue . If we follow your example with some assumed gray values:

After Image<Gray, byte> image = receivedImage.Convert<Gray, byte>().ThresholdBinary(new Gray(80), new Gray(255)); , you will produce:

This is in fact the image that you are passing to CvInvoke.FindContours() and subsequently finding only the out most contour.

What you need, if indeed you want to continue with FindContours , is an algorithm that will "bin", or "bandpass" your image to first produce the following segments, each to be converted to binary, and contour detected independently.

I feel that you currently example is probably and over simplification of the problem to offer you a solution on how you might achieve that here. However please do ask another question with more realistic data, and I will be happy to provide some suggestions.

Alternatively look towards more sophisticated edge detection methods such and Canny or Sobel. This video may be a good starting point: Edge Detection