Revision history [back]

Probably connectedComponentsWithStats is not the most efficient way to remove dots. I would rather use the erosion or the opening (erosion+dilatation) morphology operator.

If you want to stuck with connectedComponentsWithStats, the code (C++) will be:

for(int i=0;i<stats.rows;i++)    //for each label 
    if(stats.at<ushort>(i,CC_STAT_AREA)<=3)    //if area<3
        for(int y=centroids.at<float>(i,1)-1;y<=centroids.at<float>(i,1)+1;y++)   // scan the neighboring pixels
            for(int x=centroids.at<float>(i,0)-1;x<=centroids.at<float>(i,0)+1;x++)
                if(labels.at<ushort>(x,y)==i)image.at<uchar>(x,y))=0;   // if the label matches, set the current pixel to 0

I commented the code, so it will be easier to rewrite in Python.

Probably connectedComponentsWithStats is not the most efficient way to remove dots. I would rather use the erosion or the opening closing (erosion+dilatation) morphology operator.

If you want to stuck with connectedComponentsWithStats, the code (C++) will be:

for(int i=0;i<stats.rows;i++)    //for each label 
    if(stats.at<ushort>(i,CC_STAT_AREA)<=3)    //if area<3
        for(int y=centroids.at<float>(i,1)-1;y<=centroids.at<float>(i,1)+1;y++)   // scan the neighboring pixels
            for(int x=centroids.at<float>(i,0)-1;x<=centroids.at<float>(i,0)+1;x++)
                if(labels.at<ushort>(x,y)==i)image.at<uchar>(x,y))=0;   // if the label matches, set the current pixel to 0

I commented the code, so it will be easier to rewrite in Python.

Probably connectedComponentsWithStats is not the most efficient way to remove dots. I would rather use the erosion or the closing opening (erosion+dilatation) morphology operator.

If you want to stuck with connectedComponentsWithStats, the code (C++) will be:

for(int i=0;i<stats.rows;i++)    //for each label 
    if(stats.at<ushort>(i,CC_STAT_AREA)<=3)    //if area<3
        for(int y=centroids.at<float>(i,1)-1;y<=centroids.at<float>(i,1)+1;y++)   // scan the neighboring pixels
            for(int x=centroids.at<float>(i,0)-1;x<=centroids.at<float>(i,0)+1;x++)
                if(labels.at<ushort>(x,y)==i)image.at<uchar>(x,y))=0;   // if the label matches, set the current pixel to 0

I commented the code, so it will be easier to rewrite in Python.