CODEWITHSUNDEEP
c++fftwdft
dectroo
dectroo

Reputation: 161

FFTW3 DFT 2D always return NaN

I'm trying to do DFT for a real 2D matrix using FFTW3. this is my code snippet:

size_t nyh = ny/2 + 1;
out = (fftw_complex*)fftw_malloc ( sizeof ( fftw_complex ) * nx * nyh );

plan_forward = fftw_plan_dft_r2c_2d ( nx, ny, in, out, FFTW_ESTIMATE );

fftw_execute ( plan_forward );

where in is a double* each value is at order of e-270.

my problem is that out contains always -nan values.

Is there any trick to get correct values?

Thank you

Upvotes: 0

Views: 1442

Answers (1)

dectroo
dectroo

Reputation: 161

Thank everybody.

The problem which i was running in was not due to FFTW3 library and my code was correct despite of it's weakness.

My problem came from the input that i feed to fftw_plan_dft_r2c_2d function which is issued from a cv::Mat image. So, I post the error i did to help anyone getting this trouble:)

I convert cv::Mat to a vector<vector<double> > using this function:

static MatrixOfDouble _convertMat( cv::Mat& inMat)
{
    MatrixOfDouble result;

    for (int i = 0; i < inMat.rows; ++i)
    {
        std::vector<double> row;
        for (int j = 0; j < inMat.cols; ++j)
        {
            row.push_back(inMat.at<double>(i, j));
        }
        result.push_back(row);
    }

    return result;
}

And Then i liniarize the returned matrix to feed the dft function.

But some of result matrix elements was NaN.

As a fix, i get cv::Mat element as long instead of double:

row.push_back(inMat.at<long>(i, j));

Good luck for everyone trying to solve this issue.

Upvotes: 1

Related Questions