Find the median of a single channel using OpenCv

cv_median.cpp

namespace cv {
    // calculates the median value of a single channel
    // based on https://github.com/arnaudgelas/OpenCVExamples/blob/master/cvMat/Statistics/Median/Median.cpp
    double median( cv::Mat channel )
    {
        double m = (channel.rows*channel.cols) / 2;
        int bin = 0;
        double med = -1.0;

        int histSize = 256;
        float range[] = { 0, 256 };
        const float* histRange = { range };
        bool uniform = true;
        bool accumulate = false;
        cv::Mat hist;
        cv::calcHist( &channel, 1, 0, cv::Mat(), hist, 1, &histSize, &histRange, uniform, accumulate );

        for ( int i = 0; i < histSize && med < 0.0; ++i )
        {
            bin += cvRound( hist.at< float >( i ) );
            if ( bin > m && med < 0.0 )
                med = i;
        }

        return med;
    }
}

Thanks. Works fine.

Thanks for sharing. But only works for 8 bit, although it does allow any other type (with invalid results) to pass.

thanks for sharing

Good work

@TimZaman Simply change the 256 to whatever the maximum number is for the datatype + 1. So if it's unsigned 16-bit, this would be 65536 or if it's unsigned 32-bit, it would be 2^32, and so on.

@rayryeng be carefull with that, because the function isn't going to work well with all the other types. If you use int, the size of the array would be gigantic, and hence very slow.
@heisters the function is not taking into account when the matrix is even-sized, in which should use the two middle values.
Thank you!

@facug91 - I know. That's why I don't use this code when calculating the median value. I use something else instead. Just wanted to put that comment in for the sake of completeness.

@rayryeng can you tell me what you use instead?

Thanks