Matlab's ifft2 function in OpenCV

Question

I've already searched a lot on the internet but still haven't come across with a solution. What I want is to convert the Matlab's ifft2 function into OpenCV to obtain the same result. For instance, if I have a 10x10 matrix A as:

A =

    11    22    33     1     1     1     1     1     1     1
    44    55    66     1     1     1     1     1     1     1
    77    88    99     1     1     1     1     1     1     1
   100   111   122     1     1     1     1     1     1     1
     1     1     1     1     1     1     1     1     1     1
     1     1     1     1     1     1     1     1     1     1
     1     1     1     1     1     1     1     1     1     1
     1     1     1     1     1     1     1     1     1     1
     1     1     1     1     1     1     1     1     1     1
     1     1     1     1     1     1     1     1     1     1

Then in Matlab, calling ifft2(A) gives me this:

ans =

   9.1600 + 0.0000i   5.4570 + 4.6041i   0.5640 + 4.4443i  -1.1170 + 0.7295i   1.0560 - 1.4066i   2.7200 + 0.0000i   1.0560 + 1.4066i  -1.1170 - 0.7295i   0.5640 - 4.4443i   5.4570 - 4.6041i
   1.7274 + 6.5730i  -2.5500 + 5.5266i  -3.6046 + 1.6030i  -1.0629 - 0.6619i   1.2068 + 0.5078i   0.5758 + 2.1910i  -0.9060 + 1.3047i   0.2070 - 0.8463i   3.4605 - 0.4030i   4.7141 + 3.3760i
  -3.6507 + 1.1351i  -3.2357 - 1.1758i  -1.1907 - 1.8927i   0.1292 - 0.6560i  -0.3480 + 0.5913i  -1.2169 + 0.3784i  -0.8219 - 0.3575i   0.2811 + 0.1882i   0.0968 + 1.8925i  -1.8722 + 2.6341i
   0.6876 - 1.9331i   1.4480 - 0.9227i   0.9832 + 0.1199i   0.0550 + 0.1593i  -0.1955 - 0.4607i   0.2292 - 0.6444i   0.3196 - 0.1496i  -0.3841 + 0.0821i  -1.0290 - 0.6749i  -0.5821 - 1.7727i
   1.1457 + 1.6084i  -0.1392 + 1.7632i  -0.8331 + 0.7911i  -0.3628 - 0.0938i   0.3857 + 0.0000i   0.3819 + 0.5361i  -0.1268 + 0.4468i  -0.0493 - 0.2667i   0.8930 - 0.4889i   1.6339 + 0.4186i
  -1.6800 + 0.0000i  -1.1861 - 0.8618i  -0.2800 - 0.8618i   0.0661 - 0.2034i  -0.2800 + 0.2034i  -0.5600 + 0.0000i  -0.2800 - 0.2034i   0.0661 + 0.2034i  -0.2800 + 0.8618i  -1.1861 + 0.8618i
   1.1457 - 1.6084i   1.6339 - 0.4186i   0.8930 + 0.4889i  -0.0493 + 0.2667i  -0.1268 - 0.4468i   0.3819 - 0.5361i   0.3857 - 0.0000i  -0.3628 + 0.0938i  -0.8331 - 0.7911i  -0.1392 - 1.7632i
   0.6876 + 1.9331i  -0.5821 + 1.7727i  -1.0290 + 0.6749i  -0.3841 - 0.0821i   0.3196 + 0.1496i   0.2292 + 0.6444i  -0.1955 + 0.4607i   0.0550 - 0.1593i   0.9832 - 0.1199i   1.4480 + 0.9227i
  -3.6507 - 1.1351i  -1.8722 - 2.6341i   0.0968 - 1.8925i   0.2811 - 0.1882i  -0.8219 + 0.3575i  -1.2169 - 0.3784i  -0.3480 - 0.5913i   0.1292 + 0.6560i  -1.1907 + 1.8927i  -3.2357 + 1.1758i
   1.7274 - 6.5730i   4.7141 - 3.3760i   3.4605 + 0.4030i   0.2070 + 0.8463i  -0.9060 - 1.3047i   0.5758 - 2.1910i   1.2068 - 0.5078i  -1.0629 + 0.6619i  -3.6046 - 1.6030i  -2.5500 - 5.5266i

In OpenCV, I wrote the following code (thanks to internet), and indeed achieved the same result as that of Matlab.

cv::Mat inverseDFT;
cv::Mat planes[] = { cv::Mat_<float>(input), cv::Mat::zeros(input.size(), CV_32F) };
cv::Mat complex;
cv::merge(planes, 2, complex);
cv::idft(complex, inverseDFT, cv::DFT_SCALE | cv::DFT_INVERSE);

So far this is good. But, when I use another matrix of size 82x82 of same type and value, the Matlab's result is very different from that of OpenCV's. This is the Matlab output (first two rows):

1.0e+05 * 
[-0.0000   -9.6761   -4.6365    1.8979    1.4972   -0.0025    0.1037   -0.0919    0.3140    0.0032    0.0195    0.0057   -0.0002    0.0147   -0.0023    0.0088    0.0045   -0.0023    0.0135;
-0.6180    4.6026    1.9042   -1.0747   -0.5034   -0.0088   -0.0466    0.0479   -0.1339   -0.0019   -0.0078   -0.0019   -0.0031   -0.0098    0.0009   -0.0038   -0.0053    0.0034   -0.0070;
....]

And this is the OpenCV result (first two rows):

[15.8347    148.3   15.5297 148.068 15.3074 147.476 15.1266 146.854 14.9336 146.121 14.649  145.361 14.4584 144.357 14.1337 143.344 13.8938 142.052 13.4711 140.892 12.9521 139.762 12.3772 138.7   11.7153 137.773 11.149  136.736 10.6333 136.07  10.0025 135.68  9.35163 135.744 8.37026 136.085 7.51313 136.597 6.65663 137.297 5.84039 138.144 5.08981 138.964 4.2696  139.838 3.93125 140.847 3.83195 141.894 4.0224  142.883 4.54194 143.872 5.35709 144.933 6.18112 145.637 6.97564 146.306 7.91474 146.833 8.90979 147.268 9.90416 147.69  10.7855 147.929 11.9145 148.177 13.3188 148.587 14.5545 148.747 15.3029 148.96  15.8501 149.022 16.1713 149.002 16.0789 148.678 ; 
83.1579 -72.1501    83.2033 -72.3991    83.0443 -72.9689    83.028  -73.201 82.9732 -73.2787    82.9527 -73.2777    83.039  -73.2688    83.2974 -73.331 83.7781 -73.6298    84.5012 -73.8774    85.2545 -74.02  86.2663 -74.1823    87.4866 -74.3593    88.8467 -74.474 90.3635 -74.5438    91.6828 -74.5036    92.9061 -73.9815    93.1934 -73.0743    93.4254 -72.1992    93.1917 -71.2198    92.45   -70.201 91.1874 -69.4783    89.0805 -68.5612    87.2752 -67.461 85.2434 -66.3476    83.1033 -65.316 80.9829 -64.4404    79.0108 -63.7758    77.3056 -63.3485    75.7612 -63.1322    74.7189 -63.3549    74.1633 -63.9313    74.1057 -64.8886    74.2431 -66.0125    74.9995 -66.9026    76.9265 -67.4861    78.861  -68.5714    80.1996 -69.2898    81.3412 -70.1617    82.2211 -70.9372    82.7735 -71.4485;
    ....]

What has caused this difference? Could someone help on? Thanks.

What do you mean by "of same type and value"? Both your Matlab and OpenCV output for the 82x82 case seem to be real. They probably should be complex. Another thing is that your Matlab output seems to be incomplete (rows don't have enough entries). Maybe you messed up the input? — havogt, Jul 16 '15 at 17:05

score 0 · Answer 1 · answered Nov 28 '16 at 15:43

0

Actually fft in Matlab is a Discrete Fourier transform, I found that the same output can be reached via OpenCV's dct

answered Nov 28 '16 at 15:43

Abdarhman Taha

83
7

Matlab's ifft2 function in OpenCV

1 Answers1