Cfrgtkky

I am trying to implementing the inverse of Fourier transform. Here's my code:

import numpy as np 

import cv2 as cv

import math 

import cmath

from matplotlib import pyplot as plt

image="test2.bmp" 

img=cv.imread(image,cv.IMREAD_GRAYSCALE)

#foruior transform 

dft = cv.dft(np.float32(img),flags = cv.DFT_COMPLEX_OUTPUT)

#shifting for displaying 

dft_shift = np.fft.fftshift(dft)

#get the filter contains real and complex part 

t=1

a=0.1 

b=0.1

motion_filter=np.empty((img.shape[0],img.shape[1],2),dtype=np.complex_)

for x in range(img.shape[0]):

    for y in range(img.shape[1]):

        if x==0 and y==0:

            const1=math.pi*(1e-10)

        else:

            const1=math.pi*((x*a)+(y*b)) 

        #for real number

        motion_filter[x,y,0]=((t/const1)*(math.sin(const1))*(cmath.exp((0-1j)*(const1)))).real

        #for complex number

        motion_filter[x,y,1]=((t/const1)*(math.sin(const1))*(cmath.exp((0-1j)*(const1)))).imag

#processing

fshift = dft_shift*motion_filter

#shift back 

f_ishift = np.fft.ifftshift(fshift)

#inverse 

img_back = cv.idft(f_ishift)

#take real part 

img_back=img_back[:,:,0]

#show image 

plt.imshow(img_back,cmap='gray')

plt.show()

And the error occurs when doing:

img_back = cv.idft(f_ishift)

The error message is:

src data type = 15 is not supported

How can I fix the code?

According to the answer in this other question, the OpenCV idft requires a real-valued matrix where the real and imaginary components are stored along a third dimension. You create this matrix:

motion_filter=np.empty((img.shape[0],img.shape[1],2),dtype=np.complex_)

It is of the right sizes (2 along the 3rd dimension, for the real and imaginary components), but it is complex-valued. Next you multiply your Fourier-domain image (a real-valued matrix with real and imaginary components along the 3rd dimension) with this complex matrix, creating a complex-valued output:

fshift = dft_shift*motion_filter

This complex-valued output cannot be used in cv.idft. Instead, create your filter matrix as a real-valued matrix:

motion_filter=np.empty((img.shape[0],img.shape[1],2))