Properties of the Fourier Transform

One-dimensional Fourier Transformation is defined as follows:

$\label{Fourier1} H(k) = \mathcal{F}\left( h(x) \right) = \int \limits_{-\infty}^{\infty} h(x) \mathrm{e}^{-\mathrm{i}2\pi kx} \mathrm{d}x$ (1)

We mention here that the domain of this transformation is the set of functions whose absolute value has a finite integral over $\mathbb{R}$ but hereinafter we will always suppose that the picked function is element of this set as in practice we will deal with finite functions limited in a finite time interval, and therefore their absolute value can always be integrated.

The inverse Fourier transformation is very similar to the transformation itself, they only differ in a complex conjugate:

$\label{Fourier2} h(x) = \mathcal{F}^{-1}\left( H(k) \right) = \int \limits_{-\infty}^{\infty} H(k) \mathrm{e}^{\mathrm{i}2\pi kx} \mathrm{d}k$ (2)

The frequently used properties of the Fourier Transform are the followings:

1) Linearity:

$\label{Fourier3} \mathcal{F}\left( \alpha g(x) + \beta h(x) \right) = \alpha \mathcal{F} \left( g(x) \right) + \beta \mathcal{F} \left( h(x) \right)$ (3)

for every complex numbers $\alpha$ and $\beta$ . Specifically, a constant phase multiplication is preserved during the Fourier transform.

2) Shifting:

$\label{Fourier4} \mathcal{F}\left( h(x-x_0) \right) = \mathcal{F} \left( h(x) \right) \mathrm{e}^{-\mathrm{i}2 \pi k x_0} = H(k) \mathrm{e}^{-\mathrm{i}2 \pi k x_0}$ (4)

$\label{Fourier5} \mathcal{F}^{-1}\left( H(k-k_0) \right) = \mathcal{F}^{-1} \left( H(k) \right) \mathrm{e}^{\mathrm{i}2 \pi k_0 x} = h(x) \mathrm{e}^{\mathrm{i}2 \pi k_0 x}$ (5)

For every real $k_0$ and $x_0$ . This means that a linear phase ramp in the Fourier space ( $k$ -space) results a spatial shift in the image space and vice versa. The shift is proportional to the steepness of the phase ramp.

3) Convolution:

$\label{Fourier6} \mathcal{F}\left( g(x) \Conv h(x) \right) = \mathcal{F} \left( g(x) \right) \mathcal{F} \left(h(x) \right) = G(k)H(k)$ (6)

$\label{Fourier7} \mathcal{F}^{-1}\left( G(k)H(k) \right) = \mathcal{F}^{-1} \left( G(k) \right) \Conv \mathcal{F}^{-1} \left( H(k) \right) = g(x) \Conv h(x)$ (7)

$\label{Fourier8} \mathcal{F}\left( g(x) h(x) \right) = \mathcal{F} \left( g(x) \right) \Conv \mathcal{F} \left(h(x) \right) = G(k) \Conv H(k)$ (8)

$\label{Fourier9} \mathcal{F}^{-1}\left( G(k) \Conv H(k) \right) = \mathcal{F}^{-1} \left( G(k) \right) \mathcal{F}^{-1} \left( H(k) \right) = g(x) h(x)$ (9)

In other words, a multiplication in the $k$ -space results as a convolution in the image space with the inverse Fourier transformed version of the multiplying function. In the other direction, a convolution in the $k$ -space means a multiplication in the image space.

4) Symmetry. If $h(x)$ is a real function then:

$\label{Fourier10} H(-k) = \overline{H(k)}$ (10)

This symmetry is of utmost importance in the so-called partial Fourier imaging where the effective spin density is assumed to be real, and therefore the $k$ -space data is redundant as negative $k$ -vector values can be replaced by the complex conjugate of the positive $k$ -vector values.

The extension of the upper properties to higher dimensions can be done quite intuitively. The variables are then vectorial denoted by $\mathbf{r}$ and $\mathbf{k}$ , and the Fourier Transform (for example in 3D) is defined by the following with $\mathbf{kr}$ equals to their dot product:

$\label{Fourier10} H(\mathbf{k}) = \int h(\mathbf{r}) \mathrm{e}^{-\mathrm{i}2\pi \mathbf{kr}} \mathrm{d^3r}$ (11)

Export

Medical Imaging

Site Language: English

Properties of the Fourier Transform

Sidebar

Medical Imaging

Site Language: English