Revision history [back]

You might simply try to implement it yourself, without opencv functions:

Mat image, result;
    for(y=1;y<image.rows-1;y++)
        for(x=1;x<image.cols-1;x++)
            if((image.at<uchar>(x,y)>image.at<uchar>(x-1,y-1))&&(image.at<uchar>(x,y)>image.at<uchar>(x,y-1))&&(image.at<uchar>(x,y)>image.at<uchar>(x+1,y-1))&&(image.at<uchar>(x,y)>image.at<uchar>(x+1,y))&&(image.at<uchar>(x,y)>image.at<uchar>(x+1,y+1))&&(image.at<uchar>(x,y)>image.at<uchar>(x,y+1))&&(image.at<uchar>(x,y)>image.at<uchar>(x-1,y+1))&&(image.at<uchar>(x,y)>image.at<uchar>(x-1,y)))mask.at<uchar>(x,y)=1;
            else mask.at<uchar>(x,y)=0;

Sorry for the C code, it's untested and not optimized, but it's easy to understand: you take every pixel (x,y) (except at the border), and you compare it to the 8 neighbors (x-1,y-1 ; x,y-1 ; x+1,y-1 ; ...). If it's bigger, than it's a local maximum.

You can adapt and optimize the code according to your needs. With a little work you can process the borders, too.