How to Zero Pad in Image processing (FFT)

Question

I am new to image processing and this question is more theoretical.

Suppose I have a 5x5 image and a 3x3 kernel. I now want to use FFT2 to convolve the image. From what I understand you would need to pad the image with zeroes so it would look like

0000000 
0iiiii0 
0iiiii0 
0iiiii0 
0iiiii0
0iiiii0 
0000000

to accomodate for the kernel and the kernel to

0000000 
0000000 
00kkk00 
00kkk00 
00kkk00
0000000 
0000000

and im then using np.fft.fft2 to convert them both into their fourier domain versions, then multiplying and using np.fft.ifft2 to invert. After this I get imaginary values. Is this correct and should I be getting these imaginary values? If so what should I do next?

Here is the code from which I am getting imaginary values:

conv_kernel = np.zeros(image.shape)
    
for i in range(len(kernel)):
    for j in range(len(kernel[0])):
        conv_kernel[i,j] = kernel[i,j]

fourier_domain_img = np.fft.fft2(image)
fourier_domain_kernel = np.fft.fft2(conv_kernel)

multiplication_result = np.multiply(fourier_domain_img, fourier_domain_kernel)

output = np.fft.ifft2(multiplication_result)

Note that the image has been padded as mentioned previously with 0s around the edges and the kernel has been shifted to the top left and saved in conv_kernel.

output gives me imaginary values. Output snippet:

0.34701525+7.90350302e-19j 0.24993464-1.97207468e-17j
  0.25917211-7.08625583e-17j 0.26527233-7.02379773e-17j
  0.25986928-8.79273688e-17j 0.30030501-1.20158599e-16j
  0.41908497-6.28590074e-17j 0.56592593-7.59364420e-17j
  0.66858388-1.28485529e-17j 0.63956427+1.46374819e-17j
  0.52383442-2.31376484e-17j 0.30623094-7.88434060e-18j]]
complex128```