I'm making a couple of wrapper methods for fourier transform of image and have encountered a problem. Whenever I use the standard methods:
void fft::dfft(Mat* img,Mat* result){
Mat image;
img->convertTo(image, CV_32F);
cv::dft(image, *result, cv::DFT_SCALE|cv::DFT_COMPLEX_OUTPUT);
}
Invert the operations:
void fft::idft(Mat* fourier, Mat* img) {
//invert:
Mat inverseTransform;
dft(*fourier, inverseTransform, cv::DFT_INVERSE|cv::DFT_REAL_OUTPUT);
//restore image to 8 bits (fourier needs to use float at least
Mat res;
inverseTransform.convertTo(res, CV_8U);
res.copyTo(*img);
}
everything is ok. However when I try to use this pair:
void fft::dfft(Mat* img, Mat* mag, Mat* phi, Mat* re, Mat* im) {
/*SPECAL NOTE: dft is faster for images with size being multiple of two three and five. An OCV fmethod getOptrimalDFTSize comes in handy*/
Mat paddedImg; //expand input image to the optimal size
int m = getOptimalDFTSize(img->rows);
int n = getOptimalDFTSize(img->cols);
copyMakeBorder(*img, paddedImg, 0, m - img->rows, 0, n-img->cols, BORDER_CONSTANT, Scalar::all(0));
/*prepare space to store the image*/
Mat planes[] = {Mat_<float>(paddedImg), Mat::zeros(paddedImg.size(), CV_32F)};
Mat complexImg;
merge(planes, 2, complexImg);
/*Actual dft:*/
dft(complexImg, complexImg);
/*Get magnitude:*/
split(complexImg, planes);
Mat magImg, phasImg;
planes[0].copyTo(*re);
planes[1].copyTo(*im);
cout << "intern Real = "<< planes[0] << endl;
magnitude(planes[0], planes[1], magImg); //magnitude will be stored in planes[0]
phase(planes[0], planes[1], phasImg);
magImg.copyTo(*mag);
phasImg.copyTo(*phi);
#ifdef debug
namedWindow("Input Image", 0 );
imshow("Input Image", *img);
namedWindow( "Image magnitude", 0 );
imshow("Image magnitude", magImg);
waitKey(0);
#endif
}
To extract the specific parameters, I can't put it back together. Currently I'm focusing on Real and Imaginary part, because I feel that the Mag/Phi one will work in identical way just converting to real and imaginary back from Mag/Phi first.
void fft::idft(Mat* re, Mat* im, Mat* img, bool dummy) {
Mat inverseTransform;
Mat fourier(re->rows, re->cols, CV_32FC2);
vector<Mat> planes;
planes.push_back(*re);
planes.push_back(*im);
//re and im get inside correctly (verified)
Mat padded;
Mat complex;
Mat pp[] = {Mat_<float>(*re), Mat::zeros(re->size(), CV_32F)};
pp[0] = *re;
pp[1] = *im;
merge(pp, 2, complex); // Add to the expanded another plane with zeros
cv::merge(planes, fourier);
dft(complex, inverseTransform, cv::DFT_INVERSE|cv::DFT_REAL_OUTPUT);
Mat res;
inverseTransform.convertTo(res, CV_8U);
cv::imshow("iFFT1", res);
dft(fourier, inverseTransform, cv::DFT_INVERSE|cv::DFT_REAL_OUTPUT);
inverseTransform.convertTo(res, CV_8U);
cv::imshow("iFFT2", res);
res.copyTo(*img);
}
Both approaches (with vector and with array pp) produce the same white rectangle. To see if it does anything I passed
fft::idft(&re, &re, &img, true);
in such case result was identical in both windows (some random garbage but identical)
I suspect the problem lies in wrong merge procedure, but I have no other ideas how to restore the original form?
Thanks in advance for insights.
Here is my fourier domain filter, I think it should be useful:
#pragma once
#include <string>
#include <iostream>
#include "opencv2/opencv.hpp"
using namespace std;
using namespace cv;
//----------------------------------------------------------
// Recombinate quadrants
//----------------------------------------------------------
void Recomb(Mat &src,Mat &dst)
{
int cx = src.cols>>1;
int cy = src.rows>>1;
Mat tmp;
tmp.create(src.size(),src.type());
src(Rect(0, 0, cx, cy)).copyTo(tmp(Rect(cx, cy, cx, cy)));
src(Rect(cx, cy, cx, cy)).copyTo(tmp(Rect(0, 0, cx, cy)));
src(Rect(cx, 0, cx, cy)).copyTo(tmp(Rect(0, cy, cx, cy)));
src(Rect(0, cy, cx, cy)).copyTo(tmp(Rect(cx, 0, cx, cy)));
dst=tmp;
}
//----------------------------------------------------------
// Forward fft
//----------------------------------------------------------
void ForwardFFT(Mat &Src, Mat *FImg)
{
int M = getOptimalDFTSize( Src.rows );
int N = getOptimalDFTSize( Src.cols );
Mat padded;
copyMakeBorder(Src, padded, 0, M - Src.rows, 0, N - Src.cols, BORDER_CONSTANT, Scalar::all(0));
// Create complex image
// planes[0] image , planes[1] filled by zeroes
Mat planes[2] = {Mat_<float>(padded), Mat::zeros(padded.size(), CV_32F)};
Mat complexImg;
merge(planes, 2, complexImg);
dft(complexImg, complexImg,DFT_SCALE);
// After tranform we also get complex image
split(complexImg, planes);
//
planes[0] = planes[0](Rect(0, 0, planes[0].cols & -2, planes[0].rows & -2));
planes[1] = planes[1](Rect(0, 0, planes[1].cols & -2, planes[1].rows & -2));
Recomb(planes[0],planes[0]);
Recomb(planes[1],planes[1]);
FImg[0]=planes[0].clone();
FImg[1]=planes[1].clone();
}
//----------------------------------------------------------
// Inverse FFT
//----------------------------------------------------------
void InverseFFT(Mat *FImg,Mat &Dst)
{
Recomb(FImg[0],FImg[0]);
Recomb(FImg[1],FImg[1]);
Mat complexImg;
merge(FImg, 2, complexImg);
// Inverse transform
dft(complexImg, complexImg, DFT_INVERSE);
split(complexImg, FImg);
FImg[0].copyTo(Dst);
}
//----------------------------------------------------------
// Forward FFT using Magnitude and phase
//----------------------------------------------------------
void ForwardFFT_Mag_Phase(Mat &src, Mat &Mag,Mat &Phase)
{
Mat planes[2];
ForwardFFT(src,planes);
Mag.zeros(planes[0].rows,planes[0].cols,CV_32F);
Phase.zeros(planes[0].rows,planes[0].cols,CV_32F);
cv::cartToPolar(planes[0],planes[1],Mag,Phase);
}
//----------------------------------------------------------
// Inverse FFT using Magnitude and phase
//----------------------------------------------------------
void InverseFFT_Mag_Phase(Mat &Mag,Mat &Phase, Mat &dst)
{
Mat planes[2];
planes[0].create(Mag.rows,Mag.cols,CV_32F);
planes[1].create(Mag.rows,Mag.cols,CV_32F);
cv::polarToCart(Mag,Phase,planes[0],planes[1]);
InverseFFT(planes,dst);
}
//----------------------------------------------------------
// MAIN
//----------------------------------------------------------
int main(int argc, char* argv[])
{
// src image
Mat img;
// Magnitude
Mat Mag;
// Phase
Mat Phase;
// Image loading
img=imread("d:\\ImagesForTest\\lena.jpg",0);
resize(img,img,Size(512,512));
// Image size
cout<<img.size().width<<endl;
cout<<img.size().height<<endl;
//
ForwardFFT_Mag_Phase(img,Mag,Phase);
//----------------------------------------------------------
// Filter
//----------------------------------------------------------
// draw ring
int R=100; // External radius
int r=30; // internal radius
Mat mask;
mask.create(Mag.cols,Mag.rows,CV_32F);
int cx = Mag.cols>>1;
int cy = Mag.rows>>1;
mask=1;
cv::circle(mask,cv::Point(cx,cy),R,CV_RGB(0,0,0),-1);
cv::circle(mask,cv::Point(cx,cy),r,CV_RGB(1,1,1),-1);
//mask=1-mask; // uncomment for inversion
//cv::multiply(Mag,mask,Mag); // uncomment to turn filter on
//cv::multiply(Phase,mask,Phase);
//----------------------------------------------------------
// Inverse transform
//----------------------------------------------------------
InverseFFT_Mag_Phase(Mag,Phase,img);
//----------------------------------------------------------
// Results output
//----------------------------------------------------------
//
Mat LogMag;
LogMag.zeros(Mag.rows,Mag.cols,CV_32F);
LogMag=(Mag+1);
cv::log(LogMag,LogMag);
//---------------------------------------------------
imshow("Magnitude Log", LogMag);
imshow("Phase", Phase);
// img - now in CV_32FC1 format,we need CV_8UC1 or scale it by factor 1.0/255.0
img.convertTo(img,CV_8UC1);
imshow("Filtering result", img);
//----------------------------------------------------------
// Wait key press
//----------------------------------------------------------
waitKey(0);
return 0;
}
Sorry, I can't reproduce you result I've tested the code and it works fine:
#pragma once
#include <string>
#include <iostream>
#include "opencv2/opencv.hpp"
using namespace std;
using namespace cv;
//----------------------------------------------------------
// Recombinate quadrants
//----------------------------------------------------------
void Recomb(Mat &src,Mat &dst)
{
int cx = src.cols>>1;
int cy = src.rows>>1;
Mat tmp;
tmp.create(src.size(),src.type());
src(Rect(0, 0, cx, cy)).copyTo(tmp(Rect(cx, cy, cx, cy)));
src(Rect(cx, cy, cx, cy)).copyTo(tmp(Rect(0, 0, cx, cy)));
src(Rect(cx, 0, cx, cy)).copyTo(tmp(Rect(0, cy, cx, cy)));
src(Rect(0, cy, cx, cy)).copyTo(tmp(Rect(cx, 0, cx, cy)));
dst=tmp;
}
//----------------------------------------------------------
// Forward fft
//----------------------------------------------------------
void ForwardFFT(Mat &Src, Mat *FImg)
{
int M = getOptimalDFTSize( Src.rows );
int N = getOptimalDFTSize( Src.cols );
Mat padded;
copyMakeBorder(Src, padded, 0, M - Src.rows, 0, N - Src.cols, BORDER_CONSTANT, Scalar::all(0));
// Create complex image
// planes[0] image , planes[1] filled by zeroes
Mat planes[2] = {Mat_<float>(padded), Mat::zeros(padded.size(), CV_32F)};
Mat complexImg;
merge(planes, 2, complexImg);
dft(complexImg, complexImg,DFT_SCALE);
// After tranform we also get complex image
split(complexImg, planes);
//
planes[0] = planes[0](Rect(0, 0, planes[0].cols & -2, planes[0].rows & -2));
planes[1] = planes[1](Rect(0, 0, planes[1].cols & -2, planes[1].rows & -2));
Recomb(planes[0],planes[0]);
Recomb(planes[1],planes[1]);
FImg[0]=planes[0].clone();
FImg[1]=planes[1].clone();
}
//----------------------------------------------------------
// Inverse FFT
//----------------------------------------------------------
void InverseFFT(Mat *FImg,Mat &Dst)
{
Recomb(FImg[0],FImg[0]);
Recomb(FImg[1],FImg[1]);
Mat complexImg;
merge(FImg, 2, complexImg);
// Inverse transform
dft(complexImg, complexImg, DFT_INVERSE);
split(complexImg, FImg);
FImg[0].copyTo(Dst);
}
//----------------------------------------------------------
// Forward FFT using Magnitude and phase
//----------------------------------------------------------
void ForwardFFT_Mag_Phase(Mat &src, Mat &Mag,Mat &Phase)
{
Mat planes[2];
ForwardFFT(src,planes);
Mag.zeros(planes[0].rows,planes[0].cols,CV_32F);
Phase.zeros(planes[0].rows,planes[0].cols,CV_32F);
cv::cartToPolar(planes[0],planes[1],Mag,Phase);
}
//----------------------------------------------------------
// Inverse FFT using Magnitude and phase
//----------------------------------------------------------
void InverseFFT_Mag_Phase(Mat &Mag,Mat &Phase, Mat &dst)
{
Mat planes[2];
planes[0].create(Mag.rows,Mag.cols,CV_32F);
planes[1].create(Mag.rows,Mag.cols,CV_32F);
cv::polarToCart(Mag,Phase,planes[0],planes[1]);
InverseFFT(planes,dst);
}
//----------------------------------------------------------
// MAIN
//----------------------------------------------------------
int main(int argc, char* argv[])
{
// src image
Mat img;
// Magnitude
Mat Mag;
// Phase
Mat Phase;
// Image loading (grayscale)
img=imread("d:\\ImagesForTest\\lena.jpg",0);
ForwardFFT_Mag_Phase(img,Mag,Phase);
//----------------------------------------------------------
// Inverse transform
//----------------------------------------------------------
InverseFFT_Mag_Phase(Mag,Phase,img);
img.convertTo(img,CV_8UC1);
imshow("Filtering result", img);
//----------------------------------------------------------
// Wait key press
//----------------------------------------------------------
waitKey(0);
return 0;
}
And it get the same image as input.
Related
I would appreciate some help understanding the output of a small educational program I put together. I am new to OpenCV and don't have much C++ experience.
The goal of the script is to perform the following:
Load an image
Perform DFT on the image
Apply a circular binary mask to the spectrum, where the radius of the circle can be increased by hitting a key on the keyboard (essentially applying a very crude filter to the image)
Display the result of the inverse transform of the spectrum after the mask was applied
I have the basic functionality working: I can load the image, perform DFT, view the output image and increase the radius of the circle (advancing through a for-loop with the circle radius following i), and see the result of the inverse transform of the modified spectrum.
However I do not understand why the output is showing a vertically flipped copy of the input superimposed on the image (see example below with Lena.png). This is not my intended result. When I imshow() the inverse DFT result without applying the mask, I get a normal, non-doubled image. But after applying the mask, the IDFT output looks like this:
I am not looking for a quick solution: I would really appreciate if someone more experienced could ask leading questions to help me understand this result so that I can try to fix it myself.
My code:
#include <opencv2/core/core.hpp>
#include <opencv2/highgui.hpp>
#include <opencv2/imgcodecs.hpp>
#include <opencv2/imgproc.hpp>
#include <iostream>
using namespace cv;
void expand_img_to_optimal(Mat &padded, Mat &img);
void fourier_transform(Mat &img);
int main(int argc, char **argv)
{
Mat input_img;
input_img = imread("Lena.png" , IMREAD_GRAYSCALE);
if (input_img.empty())
{
fprintf(stderr, "Could not Open image\n\n");
return -1;
}
fourier_transform(input_img);
return 0;
}
void fourier_transform(Mat &img)
{
Mat padded;
expand_img_to_optimal(padded, img);
Mat planes[] = {Mat_<float>(padded), Mat::zeros(padded.size(), CV_32F)};
Mat complexI;
merge(planes, 2, complexI);
dft(complexI, complexI, DFT_COMPLEX_OUTPUT);
// For-loop to increase mask circle radius incrementally
for (int i=0; i<400; i+=10) {
Mat mask = Mat::ones(complexI.size(), complexI.type());
mask.convertTo(mask, CV_8U);
Mat dest = Mat::ones(complexI.size(), complexI.type());
circle(mask, Point(mask.cols/2, mask.rows/2), i, 0, -1, 8, 0);
complexI.copyTo(dest, mask);
Mat inverseTransform;
idft(dest, inverseTransform, DFT_INVERSE|DFT_REAL_OUTPUT);
normalize(inverseTransform, inverseTransform, 0, 1, NORM_MINMAX);
imshow("Reconstructed", inverseTransform);
waitKey(0);
}
}
void expand_img_to_optimal(Mat &padded, Mat &img) {
int row = getOptimalDFTSize(img.rows);
int col = getOptimalDFTSize(img.cols);
copyMakeBorder(img, padded, 0, row - img.rows, 0, col - img.cols, BORDER_CONSTANT, Scalar::all(0));
}
This happens because you are inverse-transforming a frequency spectrum that is not conjugate-symmetric around the origin.
The origin of the frequency domain image is the top-left pixel. Your disk mask must be centered there. The frequency domain is periodic, so that the part of the mask that extends to the left of the image wraps around and comes in to the right edge, same with top and bottom edges.
The easiest way to generate a proper mask is to
create it with the origin at (mask.cols/2, mask.rows/2), like you already do, and then
apply the ifftshift operation.
OpenCV doesn’t have a ifftshift function, this answer has code that implements the ifftshift correctly.
Firstly I'd like to thank #Cris Luengo for his helpful input on implementing the ifftshift in OpenCV.
In the end, the problem with my code was in this line:
Mat mask = Mat::ones(complexI.size(), complexI.type());
Instead of using the type of complexI, it looks like I should have used the type of img:
Mat mask = Mat::ones(complexI.size(), img.type());
Why? I'm not sure yet. Still trying to understand. Here is my complete code that is working how I intended:
#include <opencv2/core/core.hpp>
#include <opencv2/highgui.hpp>
#include <opencv2/imgcodecs.hpp>
#include <opencv2/imgproc.hpp>
#include <iostream>
using namespace cv;
void expand_img_to_optimal(Mat &padded, Mat &img);
void fourier_transform(Mat &img);
void ifft_shift(Mat &mask);
int main(int argc, char **argv)
{
Mat input_img;
input_img = imread("Lena.png" , IMREAD_GRAYSCALE);
if (input_img.empty())
{
fprintf(stderr, "Could not Open image\n\n");
return -1;
}
fourier_transform(input_img);
return 0;
}
void fourier_transform(Mat &img)
{
Mat padded;
expand_img_to_optimal(padded, img);
Mat planes[] = {Mat_<float>(padded), Mat::zeros(padded.size(), CV_32F)};
Mat complexI;
merge(planes, 2, complexI);
dft(complexI, complexI, DFT_COMPLEX_OUTPUT);
for (float i=0; i<4000; i+=2) {
// Create disk mask matrix
Mat mask = Mat::ones(complexI.size(), CV_8U);
circle(mask, Point(mask.cols/2, mask.rows/2), i, 0, -1, 8, 0);
// Perform ifft shift
ifft_shift(mask);
// Destination matrix for masked spectrum
Mat dest;
complexI.copyTo(dest, mask);
// Perform inverse DFT
Mat inverseTransform;
idft(dest, inverseTransform, DFT_INVERSE|DFT_REAL_OUTPUT);
normalize(inverseTransform, inverseTransform, 0, 1, NORM_MINMAX);
imshow("Reconstructed", inverseTransform);
waitKey(0);
}
}
void expand_img_to_optimal(Mat &padded, Mat &img) {
int row = getOptimalDFTSize(img.rows);
int col = getOptimalDFTSize(img.cols);
copyMakeBorder(img, padded, 0, row - img.rows, 0, col - img.cols, BORDER_CONSTANT, Scalar::all(0));
}
void ifft_shift(Mat &mask) {
// input sizes
int sx = mask.cols;
int sy = mask.rows;
// input origin
int cx = sx / 2;
int cy = sy / 2;
// split the quadrants
Mat top_left(mask, Rect(0, 0, cx, cy));
Mat top_right(mask, Rect(cx, 0, sx - cx, cy));
Mat bottom_left(mask, Rect(0, cy, cx, sy - cy));
Mat bottom_right(mask, Rect(cx, cy, sx - cx, sy - cy));
// merge the quadrants in right order
Mat tmp1, tmp2;
hconcat(bottom_right, bottom_left, tmp1);
hconcat(top_right, top_left, tmp2);
vconcat(tmp1, tmp2, mask);
}
Basically I am trying to implement a very basic version of the Wiener filter on a grey scale image, using the a stripped down Wiener equation: (1/(SNR))*DFT(Image) after which I take the IDFT of the whole thing. My problem is that my output image which is supposed to be filtered looks exactly like the input, and therefore it seems that the pixel values aren't changing at all. Can anyone please indicate to me where I am going wrong? Here's the code I am currently using:
#include "opencv2/imgproc/imgproc.hpp"
#include "opencv2/highgui/highgui.hpp"
#include "opencv/cv.hpp"
#include "opencv/cxcore.hpp"
#include <iostream>
using namespace cv;
using namespace std;
void updateMag(Mat complex);
Mat updateResult(Mat complex);
Mat computeDFT(Mat image);
Mat DFT2(Mat I);
void shift(Mat magI);
int kernel_size = 0;
int main( int argc, char** argv )
{
Mat result;
String file;
file = " << SAMPLE FILE >>";
Mat image = imread("/Users/John/Desktop/house.png", CV_LOAD_IMAGE_GRAYSCALE);
namedWindow( "Orginal window", CV_WINDOW_AUTOSIZE );// Create a window for display.
imshow( "Orginal window", image ); // Show our image inside it.
float x = 1/0.001;
Mat complex = computeDFT(image); // DFT of image
updateMag(complex); // compute magnitude of complex, switch to logarithmic scale and display...
Mat fourierImage(complex.size(), complex.type());
fourierImage = cv::Scalar::all(x);
//cout<< "Fourier = " << endl << fourierImage << endl;
//Mat complexFourier = computeDFT(fourierImage);
//cout << "1" << endl << complexFourier.type() << endl << complexFourier.type() << endl;
//complex = complex.mul(fourierImage);
//mulSpectrums(complex, fourierImage, complex, DFT_ROWS);
complex = complex.mul(x);
result = updateResult(complex); // do inverse transform and display the result image
waitKey(0);
return 0;
}
Mat updateResult(Mat complex)
{
Mat work;
//work.convertTo(work, CV_32F);
idft(complex, work);
//dft(complex, work, DFT_INVERSE + DFT_SCALE);
Mat planes[] = {Mat::zeros(complex.size(), complex.type()), Mat::zeros(complex.size(), complex.type())};
split(work, planes); // planes[0] = Re(DFT(I)), planes[1] = Im(DFT(I))
magnitude(planes[0], planes[1], work); // === sqrt(Re(DFT(I))^2 + Im(DFT(I))^2)
normalize(work, work, 1, 0, NORM_MINMAX);
imshow("result", work);
return work;
}
void updateMag(Mat complex )
{
Mat magI;
Mat planes[] = {Mat::zeros(complex.size(), CV_32F), Mat::zeros(complex.size(), CV_32F)};
split(complex, planes); // planes[0] = Re(DFT(I)), planes[1] = Im(DFT(I))
magnitude(planes[0], planes[1], magI); // sqrt(Re(DFT(I))^2 + Im(DFT(I))^2)
// switch to logarithmic scale: log(1 + magnitude)
magI += Scalar::all(1);
log(magI, magI);
shift(magI);
normalize(magI, magI, 1, 0, NORM_INF); // Transform the matrix with float values into a
// viewable image form (float between values 0 and 1).
imshow("spectrum", magI);
}
Mat computeDFT(Mat image) {
Mat padded; //expand input image to optimal size
int m = getOptimalDFTSize( image.rows );
int n = getOptimalDFTSize( image.cols ); // on the border add zero values
copyMakeBorder(image, padded, 0, m - image.rows, 0, n - image.cols, BORDER_CONSTANT, Scalar::all(0));
Mat planes[] = {Mat_<float>(padded), Mat::zeros(padded.size(), CV_32F)};
Mat complex;
merge(planes, 2, complex); // Add to the expanded another plane with zeros
dft(complex, complex, DFT_COMPLEX_OUTPUT); // furier transform
return complex;
}
void shift(Mat magI) {
// crop if it has an odd number of rows or columns
magI = magI(Rect(0, 0, magI.cols & -2, magI.rows & -2));
int cx = magI.cols/2;
int cy = magI.rows/2;
Mat q0(magI, Rect(0, 0, cx, cy)); // Top-Left - Create a ROI per quadrant
Mat q1(magI, Rect(cx, 0, cx, cy)); // Top-Right
Mat q2(magI, Rect(0, cy, cx, cy)); // Bottom-Left
Mat q3(magI, Rect(cx, cy, cx, cy)); // Bottom-Right
Mat tmp; // swap quadrants (Top-Left with Bottom-Right)
q0.copyTo(tmp);
q3.copyTo(q0);
tmp.copyTo(q3);
q1.copyTo(tmp); // swap quadrant (Top-Right with Bottom-Left)
q2.copyTo(q1);
tmp.copyTo(q2);
}
Mat DFT2(Mat I)
{
Mat padded; //expand input image to optimal size
int m = getOptimalDFTSize( I.rows );
int n = getOptimalDFTSize( I.cols ); // on the border add zero values
copyMakeBorder(I, padded, 0, m - I.rows, 0, n - I.cols, BORDER_CONSTANT, Scalar::all(0));
Mat planes[] = {Mat_<float>(padded), Mat::zeros(padded.size(), CV_32F)};
Mat complexI;
merge(planes, 2, complexI); // Add to the expanded another plane with zeros
dft(complexI, complexI); // this way the result may fit in the source matrix
// compute the magnitude and switch to logarithmic scale
// => log(1 + sqrt(Re(DFT(I))^2 + Im(DFT(I))^2))
split(complexI, planes); // planes[0] = Re(DFT(I), planes[1] = Im(DFT(I))
magnitude(planes[0], planes[1], planes[0]);// planes[0] = magnitude
Mat magI = planes[0];
magI += Scalar::all(1); // switch to logarithmic scale
log(magI, magI);
// crop the spectrum, if it has an odd number of rows or columns
magI = magI(Rect(0, 0, magI.cols & -2, magI.rows & -2));
// rearrange the quadrants of Fourier image so that the origin is at the image center
int cx = magI.cols/2;
int cy = magI.rows/2;
Mat q0(magI, Rect(0, 0, cx, cy)); // Top-Left - Create a ROI per quadrant
Mat q1(magI, Rect(cx, 0, cx, cy)); // Top-Right
Mat q2(magI, Rect(0, cy, cx, cy)); // Bottom-Left
Mat q3(magI, Rect(cx, cy, cx, cy)); // Bottom-Right
Mat tmp; // swap quadrants (Top-Left with Bottom-Right)
q0.copyTo(tmp);
q3.copyTo(q0);
tmp.copyTo(q3);
q1.copyTo(tmp); // swap quadrant (Top-Right with Bottom-Left)
q2.copyTo(q1);
tmp.copyTo(q2);
normalize(magI, magI, 0, 1, CV_MINMAX); // Transform the matrix with float values into a
// viewable image form (float between values 0 and 1).
return complexI;
}
I am using Visual Studio 2013 and OpenCV library. I display discrete Fourier transform and it works well. I'd like to save the displayed image which is the spectrum, but the saved one is black even when I use JPEG quality=100.
#include "opencv2/core/core.hpp"
#include "opencv2/imgproc/imgproc.hpp"
#include "opencv2/highgui/highgui.hpp"
#include "opencv2/opencv.hpp"
#include <iostream>
#include <cmath>
using namespace cv;
using namespace std;
static void help(char* progName)
{
cout << endl
<< "This program demonstrated the use of the discrete Fourier transform (DFT). " << endl
<< "The dft of an image is taken and it's power spectrum is displayed." << endl
<< "Usage:" << endl
<< progName << " [image_name -- default lena.jpg] " << endl << endl;
}
int main(int argc, char ** argv)
{
help(argv[0]);
const char* filename = argc >= 2 ? argv[1] : "lena.jpg";
Mat I = imread(filename, CV_LOAD_IMAGE_GRAYSCALE);
if (I.empty())
return -1;
Mat padded; //expand input image to optimal size
int m = getOptimalDFTSize(I.rows);
int n = getOptimalDFTSize(I.cols); // on the border add zero values
copyMakeBorder(I, padded, 0, m - I.rows, 0, n - I.cols, BORDER_CONSTANT, Scalar::all(0));
Mat planes[] = { Mat_<float>(padded), Mat::zeros(padded.size(), CV_32F) };
Mat complexI;
merge(planes, 2, complexI); // Add to the expanded another plane with zeros
dft(complexI, complexI); // this way the result may fit in the source matrix
// compute the magnitude and switch to logarithmic scale
// => log(1 + sqrt(Re(DFT(I))^2 + Im(DFT(I))^2))
split(complexI, planes); // planes[0] = Re(DFT(I), planes[1] = Im(DFT(I))
magnitude(planes[0], planes[1], planes[0]);// planes[0] = magnitude
Mat magI = planes[0];
magI += Scalar::all(1); // switch to logarithmic scale
log(magI, magI);
// crop the spectrum, if it has an odd number of rows or columns
magI = magI(Rect(0, 0, magI.cols & -2, magI.rows & -2));
// rearrange the quadrants of Fourier image so that the origin is at the image center
int cx = magI.cols / 2;
int cy = magI.rows / 2;
Mat q0(magI, Rect(0, 0, cx, cy)); // Top-Left - Create a ROI per quadrant
Mat q1(magI, Rect(cx, 0, cx, cy)); // Top-Right
Mat q2(magI, Rect(0, cy, cx, cy)); // Bottom-Left
Mat q3(magI, Rect(cx, cy, cx, cy)); // Bottom-Right
Mat tmp; // swap quadrants (Top-Left with Bottom-Right)
q0.copyTo(tmp);
q3.copyTo(q0);
tmp.copyTo(q3);
q1.copyTo(tmp); // swap quadrant (Top-Right with Bottom-Left)
q2.copyTo(q1);
tmp.copyTo(q2);
normalize(magI, magI, 0, 1, CV_MINMAX); // Transform the matrix with float values into a
// viewable image form (float between values 0 and 1).
imshow("Input Image", I); // Show the result
imshow("spectrum magnitude", magI);
cv::Mat gs_bgr(magI.size(), CV_8UC1);
cv::cvtColor(magI, gs_bgr, CV_RGB2GRAY);
imwrite("orig.png", gs_bgr);
waitKey();
return 0;
}
the outcome of the dft (magI) is a float Mat, but you can only save uchar images with imwrite.
since you normalized the image to [0..1] , the resulting uchar - grayscale img only has 0 and 1 values, which will look pretty black indeed.
also, applying cv::cvtColor(magI, gs_bgr, CV_RGB2GRAY); to a 1chan float img seems broken.
instead of that, try:
Mat gray;
magI.convertTo(gray, CV_8U, 255); // upscale to [0..255]
I want to show that phase of an image carries more information than that of its magnitude, so I want to exchange the magnitude of two image and then do the inverse DFT.
here is the code:
void main()
{
Mat I1 = imread("lena.jpg", CV_LOAD_IMAGE_GRAYSCALE);
Mat I2 = imread("peppers.png", CV_LOAD_IMAGE_GRAYSCALE);
Mat padded1,padded2;
//expand input image to optimal size
int m1= getOptimalDFTSize( I1.rows );
int n1 = getOptimalDFTSize( I1.cols );
int m2= getOptimalDFTSize( I2.rows );
int n2 = getOptimalDFTSize( I2.cols );
// on the border add zero values
copyMakeBorder(I1, padded1, 0, m1 - I1.rows, 0, n1 - I1.cols, BORDER_CONSTANT, Scalar::all(0));
copyMakeBorder(I2, padded2, 0, m2 - I2.rows, 0, n2 - I2.cols, BORDER_CONSTANT, Scalar::all(0));
Mat planes1[] = {Mat_<float>(padded1), Mat::zeros(padded1.size(), CV_32F)};
Mat planes2[] = {Mat_<float>(padded2), Mat::zeros(padded2.size(), CV_32F)};
Mat complexI, complexII;
// Add to the expanded another plane with zeros
merge(planes1, 2, complexI);
merge(planes2, 2, complexII);
dft(complexI, complexI);
dft(complexII, complexII);
// compute the magnitude and phase then switch to logarithmic scale
// => magnitude:log(1 + sqrt(Re(DFT(I))^2 + Im(DFT(I))^2)), phase:arctan(Im(DFT(I)),Re(DFT(I)))
split(complexI, planes1);// planes[0] = Re(DFT(I)), planes[1] = Im(DFT(I))
Mat ph1, magI1;
phase(planes1[0], planes1[1], ph1);//ph1 = phase
magnitude(planes1[0], planes1[1], magI1);// magI1 = magnitude
magI1 = magI1(Rect(0, 0, magI1.cols & -2, magI1.rows & -2));
ph1 = ph1(Rect(0, 0, ph1.cols & -2, ph1.rows & -2));
split(complexII, planes2);// planes[0] = Re(DFT(I)), planes[1] = Im(DFT(I))
Mat ph2, magI2;
phase(planes2[0], planes2[1], ph2);//ph2 = phase
magnitude(planes2[0], planes2[1], magI2);// mag2 = magnitude
magI2 = magI2(Rect(0, 0, magI2.cols & -2, magI2.rows & -2));
ph2 = ph2(Rect(0, 0, ph2.cols & -2, ph2.rows & -2));
planes1[1] = ph1; planes1[0] = magI2;
planes2[1] = ph2; planes2[0] = magI1;
dft(complexI,complexI,DFT_INVERSE);
dft(complexII,complexII,DFT_INVERSE);
imshow("image", complexI);
waitKey();
}
I just simply merge magnitude and phase together then do the IDFT, seems totally wrong.
I guess from your question that something with your dft is not working.
Try the below Code (after adding your split Planes) and see if it works.
The images have to be the Exact same Size.
If something else is wrong: please show your images and results. Maybe your code is correct and you are just expecting the wrong thing.
Here is the working Example:
// Load an image
Mat I1 = imread("lena.jpg", CV_LOAD_IMAGE_GRAYSCALE);
Mat I2 = imread("peppers.png", CV_LOAD_IMAGE_GRAYSCALE);
Mat fI1;
Mat fI2;
I1.convertTo(fI1, CV_32F);
I2.convertTo(fI2, CV_32F);
//Perform DFT
Mat fourierTransform1;
Mat fourierTransform2;
dft(fI1, fourierTransform1, DFT_SCALE|DFT_COMPLEX_OUTPUT);
dft(fI2, fourierTransform2, DFT_SCALE|DFT_COMPLEX_OUTPUT);
//your split plane and everything else
//Perform IDFT
Mat inverseTransform1;
Mat inverseTransform2;
dft(fourierTransform1, inverseTransform1, DFT_INVERSE|DFT_REAL_OUTPUT);
dft(fourierTransform2, inverseTransform2, DFT_INVERSE|DFT_REAL_OUTPUT);
Mat result1;
Mat result2;
inverseTransform1.convertTo(result1, CV_8U);
inverseTransform2.convertTo(result2, CV_8U);
You can't merge the magnitude and phase by using cv::merge() function, instead you should use cv::polarToCart(). Here's my code to do what you want:
using namespace cv;
// Rearrange the quadrants of a Fourier image so that the origin is at
// the image center
void shiftDFT(Mat &fImage )
{
Mat tmp, q0, q1, q2, q3;
// first crop the image, if it has an odd number of rows or columns
fImage = fImage(Rect(0, 0, fImage.cols & -2, fImage.rows & -2));
int cx = fImage.cols / 2;
int cy = fImage.rows / 2;
// rearrange the quadrants of Fourier image
// so that the origin is at the image center
q0 = fImage(Rect(0, 0, cx, cy));
q1 = fImage(Rect(cx, 0, cx, cy));
q2 = fImage(Rect(0, cy, cx, cy));
q3 = fImage(Rect(cx, cy, cx, cy));
q0.copyTo(tmp);
q3.copyTo(q0);
tmp.copyTo(q3);
q1.copyTo(tmp);
q2.copyTo(q1);
tmp.copyTo(q2);
}
int main()
{
// Load an image
Mat I1 = imread("lena.jpg", CV_LOAD_IMAGE_GRAYSCALE);
Mat I2 = imread("pepper.jpg", CV_LOAD_IMAGE_GRAYSCALE);
Mat fI1;
Mat fI2;
I1.convertTo(fI1, CV_32F);
I2.convertTo(fI2, CV_32F);
//expand input image to optimal size
int m = getOptimalDFTSize( I1.rows );
int n = getOptimalDFTSize( I1.cols );
Mat padded1, padded2;
// on the border add zero values
copyMakeBorder(fI1, padded1, 0, m - I1.rows, 0, n - I1.cols, BORDER_CONSTANT, Scalar::all(0));
copyMakeBorder(fI2, padded2, 0, m - I2.rows, 0, n - I2.cols, BORDER_CONSTANT, Scalar::all(0));
//Perform DFT
Mat fourierTransform1;
Mat fourierTransform2;
Mat planes1[2], planes2[2];
dft(fI1, fourierTransform1, DFT_SCALE|DFT_COMPLEX_OUTPUT);
dft(fI2, fourierTransform2, DFT_SCALE|DFT_COMPLEX_OUTPUT);
shiftDFT(fourierTransform1);
shiftDFT(fourierTransform2);
split(fourierTransform1, planes1);// planes[0] = Re(DFT(I)), planes[1] = Im(DFT(I))
split(fourierTransform2, planes2);// planes[0] = Re(DFT(I)), planes[1] = Im(DFT(I))
Mat ph1, mag1;
mag1.zeros(planes1[0].rows, planes1[0].cols, CV_32F);
ph1.zeros(planes1[0].rows, planes1[0].cols, CV_32F);
cartToPolar(planes1[0], planes1[1], mag1, ph1);
Mat ph2, mag2;
mag2.zeros(planes2[0].rows, planes2[0].cols, CV_32F);
ph2.zeros(planes2[0].rows, planes2[0].cols, CV_32F);
cartToPolar(planes2[0], planes2[1], mag2, ph2);
polarToCart(mag1, ph2, planes1[0], planes1[1]);
polarToCart(mag2, ph1, planes2[0], planes2[1]);
merge(planes1, 2, fourierTransform1);
merge(planes2, 2, fourierTransform2);
shiftDFT(fourierTransform1);
shiftDFT(fourierTransform2);
//Perform IDFT
Mat inverseTransform1, inverseTransform2;
dft(fourierTransform1, inverseTransform1, DFT_INVERSE|DFT_REAL_OUTPUT);
dft(fourierTransform2, inverseTransform2, DFT_INVERSE|DFT_REAL_OUTPUT);
namedWindow("original image 1");
imshow("original image 1", I1);
namedWindow("original image 2");
imshow("original image 2", I2);
waitKey(0);
cv::Mat out1, out2;
inverseTransform1.convertTo(out1, CV_8U);
inverseTransform2.convertTo(out2, CV_8U);
namedWindow("result image 1");
imshow("result image 1", out1);
namedWindow("result image 2");
imshow("result image 2", out2);
waitKey(0);
}
This question already has an answer here:
Inverse fourier transformation in OpenCV
(1 answer)
Closed 6 years ago.
I'm trying to implement inverse DFT using OpenCV in C++
I downloaded complete dft example in docs.opencv.org and just adjust couple of lines to inverse.
my DFT code is like this
Mat DFT(const char* filename)
{
Mat I = imread(filename, CV_LOAD_IMAGE_GRAYSCALE);
if (I.empty())
{
Mat emty(7, 7, CV_32FC2, Scalar(1, 3));
return emty;
}
Mat padded; //expand input image to optimal size
int m = getOptimalDFTSize(I.rows);
int n = getOptimalDFTSize(I.cols); // on the border add zero values
copyMakeBorder(I, padded, 0, m - I.rows, 0, n - I.cols, BORDER_CONSTANT, Scalar::all(0));
Mat planes[] = { Mat_<float>(padded), Mat::zeros(padded.size(), CV_32F) };
Mat complexI;
merge(planes, 2, complexI); // Add to the expanded another plane with zeros
dft(complexI, complexI); // this way the result may fit in the source matrix
// compute the magnitude and switch to logarithmic scale
// => log(1 + sqrt(Re(DFT(I))^2 + Im(DFT(I))^2))
split(complexI, planes); // planes[0] = Re(DFT(I), planes[1] = Im(DFT(I))
magnitude(planes[0], planes[1], planes[0]);// planes[0] = magnitude
Mat magI = planes[0];
magI += Scalar::all(1); // switch to logarithmic scale
log(magI, magI);
normalize(magI, magI, 0, 1, CV_MINMAX); // Transform the matrix with float values into a
// viewable image form (float between values 0 and 1).
imshow("Input Image", I); // Show the result
imshow(filename, magI);
// waitKey();
return magI;
}
and did IDFT. By just fix dft to idft. But the output was just look like noise. What did I do wrong? I thought the dft and idft is just same....
Mat IDFT(Mat src)
{
Mat I = src;
Mat padded; //expand input image to optimal size
int m = getOptimalDFTSize(I.rows);
int n = getOptimalDFTSize(I.cols); // on the border add zero values
copyMakeBorder(I, padded, 0, m - I.rows, 0, n - I.cols, BORDER_CONSTANT, Scalar::all(0));
Mat planes[] = { Mat_<float>(padded), Mat::zeros(padded.size(), CV_32F) };
Mat complexI;
merge(planes, 2, complexI); // Add to the expanded another plane with zeros
dft(complexI, complexI, DFT_INVERSE); // this way the result may fit in the source matrix
// compute the magnitude and switch to logarithmic scale
// => log(1 + sqrt(Re(IDFT(I))^2 + Im(IDFT(I))^2))
split(complexI, planes); // planes[0] = Re(IDFT(I), planes[1] = Im(IDFT(I))
magnitude(planes[0], planes[1], planes[0]);// planes[0] = magnitude
Mat magI = planes[0];
magI += Scalar::all(1); // switch to logarithmic scale
log(magI, magI);
normalize(magI, magI, 0, 1, CV_MINMAX);
imshow("forged map", magI);
return magI;
}
you have to rewrite your code like this to get the inverse DFT which is the original image read :
#include "stdafx.h"
#include <opencv2/core/core.hpp>
#include <opencv2\highgui\highgui.hpp>
#include <opencv2/imgproc/imgproc.hpp>
#include <iostream>
using namespace std;
using namespace cv;
int main()
{
Mat I = imread("test.tif", CV_LOAD_IMAGE_GRAYSCALE);
if( I.empty())
return -1;
Mat padded; //expand input image to optimal size
int m = getOptimalDFTSize( I.rows );
int n = getOptimalDFTSize( I.cols ); // on the border add zero values
copyMakeBorder(I, padded, 0, m - I.rows, 0, n - I.cols, BORDER_CONSTANT, Scalar::all(0));
Mat planes[] = {Mat_<float>(padded), Mat::zeros(padded.size(), CV_32F)};
Mat complexI;
merge(planes, 2, complexI); // Add to the expanded another plane with zeros
dft(complexI, complexI); // this way the result may fit in the source matrix
// compute the magnitude and switch to logarithmic scale
// => log(1 + sqrt(Re(DFT(I))^2 + Im(DFT(I))^2))
split(complexI, planes); // planes[0] = Re(DFT(I), planes[1] = Im(DFT(I))
magnitude(planes[0], planes[1], planes[0]);// planes[0] = magnitude
Mat magI = planes[0];
magI += Scalar::all(1); // switch to logarithmic scale
log(magI, magI);
// crop the spectrum, if it has an odd number of rows or columns
magI = magI(Rect(0, 0, magI.cols & -2, magI.rows & -2));
// rearrange the quadrants of Fourier image so that the origin is at the image center
int cx = magI.cols/2;
int cy = magI.rows/2;
Mat q0(magI, Rect(0, 0, cx, cy)); // Top-Left - Create a ROI per quadrant
Mat q1(magI, Rect(cx, 0, cx, cy)); // Top-Right
Mat q2(magI, Rect(0, cy, cx, cy)); // Bottom-Left
Mat q3(magI, Rect(cx, cy, cx, cy)); // Bottom-Right
Mat tmp; // swap quadrants (Top-Left with Bottom-Right)
q0.copyTo(tmp);
q3.copyTo(q0);
tmp.copyTo(q3);
q1.copyTo(tmp); // swap quadrant (Top-Right with Bottom-Left)
q2.copyTo(q1);
tmp.copyTo(q2);
normalize(magI, magI, 0, 1, CV_MINMAX); // Transform the matrix with float values into a
normalize(phaseVals, phaseVals, 0, 1, CV_MINMAX);
// viewable image form (float between values 0 and 1).
imshow("Input Image" , I ); // Show the result
imshow("Spectrum Magnitude", magI);
waitKey();
//calculating the idft
cv::Mat inverseTransform;
cv::dft(complexI, inverseTransform, cv::DFT_INVERSE|cv::DFT_REAL_OUTPUT);
normalize(inverseTransform, inverseTransform, 0, 1, CV_MINMAX);
imshow("Reconstructed", inverseTransform);
waitKey();
return 0;
}
I just added this part to your code :
//calculating the idft
cv::Mat inverseTransform;
cv::dft(complexI, inverseTransform, cv::DFT_INVERSE|cv::DFT_REAL_OUTPUT);
normalize(inverseTransform, inverseTransform, 0, 1, CV_MINMAX);
imshow("Reconstructed", inverseTransform);
waitKey();
for some reason using the normalize function in the given idft block doesn't give exactly the desired original image (contrast wise). Instead use convertTo. To be precise, replace the normalize function with the following line.
inverseTransform.convertTo(inverseTransform, CV_8U);