# Importance of Phase in FFT of an image

While processing digital images in the Fourier domain, mostly we exploit the amplitude and not the phase. This could be because the amplitude is much more structured and the amplitude spectrum reveals a lot about the spatial domain as illustrated in HIPR and other answers.

I am really curious to know what does the phase encapsulates, what can we infer from it, what kinds of operations can we perform on the phase of FFT of an image to achieve some objectives?

My specific interest lies in fusing two images in the frequency domain, but any PDF or well-documented links on understanding the phase of FFT(image) will help.

------ Edit (following images as suggested by TimWescott in comments)

• The phase contains half of all information in the image. Do a mental experiment: multiplying your whole image's FFT with $\sqrt{-1}$ does nothing to its amplitude, but what happens to the image. Now, imagine multiplying each element of the FFT with a random complex factor of magnitude 1 (i.e. not changing the amplitude, but the phase), thereby completely removing the original phase information. What's left of your image? – Marcus Müller Jan 18 at 15:36
• Better than a mental experiment: grab an image, and get it into your image processing tool. Take it's 2D FFT. Now calculate $x_{mag} = |x|$, where $|\cdot|$ is the magnitude of each FFT bin. Then calculate $x_\theta = x / x_{mag}$ (again, point-by-point). Then display the IFFT of $x_{mag}$ and $x_\theta$ as images. See what you think. – TimWescott Jan 18 at 18:42
• @TimWescott I have added the images suggested by you in the question, but still cannot understand what is that phase is encapsulating other that without it cannot recover the original image. – Mohit Lamba Jan 22 at 13:18

## 1 Answer

The amplitude of an image represents the intensity of the different frequencies in the image. Therefore, it holds the geometrical structure of features in the image (i.e. changes in the spatial domain).

The phase on the other hand, represents the locations of these features (which helps our human eye to better comprehend the image).

Here's a visualization I made in Matlab:

As you can see at the bottom, when we combine amplitude and phase from different images and perform inverse FFT, the resulting image looks much more like the original from which we took the phase. This means that phase holds more information about the image than amplitude.