Htykuut

I am using EmguCV to do simple image processing. I want to remove small contours from my thresholded image so I used connectedcomponentswithstats. Below is my code where I convert any pixel smaller than the minimum size into black - essentially deleting it. The reason I want to do this is to be able to use it in the latter part since it is not compatible with other types.

public Mat Filtered(Mat in_img, string currentDir)

{

    Mat gray_res = new Mat();

    Mat labels = new Mat();

    Mat stats = new Mat();

    Mat centroids = new Mat();



    CvInvoke.CvtColor(in_img, gray_res, ColorConversion.Bgr2Gray);



    var nlabels = CvInvoke.ConnectedComponentsWithStats(gray_res, labels, stats, centroids, LineType.EightConnected);



    int min_size = 250;



    int sizes = new int[nlabels];



    for (int s = 1; s < nlabels; s++)

    {

        sizes[s] = BitConverter.ToInt32(stats.GetData(s, (int)ConnectedComponentsTypes.Area), 0);

    }



    Matrix<double> fltr_img = new Matrix<double>(labels.Rows, labels.Cols);

    fltr_img.SetValue(0);



    for (int i = 1; i < nlabels; i++)

    {

        int j = BitConverter.ToInt32(stats.GetData(i, (int)ConnectedComponentsTypes.Left), 0); //left most

        int w = BitConverter.ToInt32(stats.GetData(i, (int)ConnectedComponentsTypes.Width), 0); //run width

        int k = BitConverter.ToInt32(stats.GetData(i, (int)ConnectedComponentsTypes.Top), 0); //top most

        int h = BitConverter.ToInt32(stats.GetData(i, (int)ConnectedComponentsTypes.Height), 0); //run height



        if (sizes[i] < min_size)

        {

            for (int a = k; a < k + h; a++)

            {

                for (int b = j; b < j + w; b++)

                {

                    fltr_img.Data[a, b] = 255;

                }

            }

        }

    }



    //Covnert 2D Matrix to Image

    CvInvoke.Imwrite(currentDir + "filter.jpg", fltr_img);



    Mat subtrahend = CvInvoke.Imread(currentDir + "filter.jpg");



    Mat fnl_img = new Mat();



    CvInvoke.Subtract(in_img, subtrahend, fnl_img);



    return fnl_img;

}

What I did, in the end, is a little dirty trick to convert the mask into a compatible type Mat which is the original image. With this code, I am getting my result but I am still bothered by it.

You can convert Matrix type by using its Mat property. I don't know what type you need so I used DepthType.Cv8U in my example.

Mat converted = new Mat();

fltr_img.Mat.ConvertTo(converted, DepthType.Cv8U);

Mat fnl_img = new Mat();

CvInvoke.Subtract(in_img, converted, fnl_img);

Alternatively, you can also use Mat and/or Matrix together with substract method. Both implement IInpurArray interface.

CvInvoke.Subtract(in_img,  fltr_img, fnl_img); //Tested with EmguCv 3.4.1