Scribd
Upload
101 views21 pages

Cpe 584 Digital Image Processing: Fourier Transform (Iv)

The document discusses the 2D continuous and discrete Fourier transforms. It explains that the basis functions of the 2D continuous Fourier transform are sinusoids with frequencies u and v in the two directions. It also describes the computational complexity of the 2D discrete Fourier transform and how it can be separated into 1D transforms along rows and columns for improved efficiency. Finally, it provides examples of 2D Fourier transforms calculated on images and discusses the relationship between image content and spatial frequencies.

Uploaded by

ali ALFODEH
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
101 views21 pages

Cpe 584 Digital Image Processing: Fourier Transform (Iv)

The document discusses the 2D continuous and discrete Fourier transforms. It explains that the basis functions of the 2D continuous Fourier transform are sinusoids with frequencies u and v in the two directions. It also describes the computational complexity of the 2D discrete Fourier transform and how it can be separated into 1D transforms along rows and columns for improved efficiency. Finally, it provides examples of 2D Fourier transforms calculated on images and discusses the relationship between image content and spatial frequencies.

Uploaded by

ali ALFODEH
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 21

CpE 584

Digital Image Processing

Fourier Transform
(IV)

CpE 584 Al-Omari


2-D Continuous Fourier Transform

Basis functions are sinusoids with frequency u


in one direction times sinusoids with frequency v
in the other.

F (u ,v ) = f ( x, y ) e
j2 (ux + v y )
dx dy

CpE 584 Al-Omari


2-D Continuous Fourier Transform

Same process for the 2-D inverse Fourier


Transform


f ( x, y ) = F(u,v )e
j2 (ux + v y )
du dv

CpE 584 Al-Omari


2-D Discrete Fourier Transform

For an NxM image, the basis functions are


j2 u x / N j2 v y / M
e e
Henceforth, the 2-D Discrete Fourier Transform
N 1 M 1 j2 u x + v y
F [u ,v ] = f [x, y ]e
1 N M

MN x =0 y =0

CpE 584 Al-Omari


2-D Discrete Fourier Transform

and, the inverse is


N 1 M 1 j2 ux + v y
f [x, y ] = F [x, y ]e N M

u =0 v =0

What is the computational complexity?

CpE 584 Al-Omari


Properties

All other properties from one-dimensional


signals apply:
Linearity
Shift
Scaling
Rayleighs Theorem
Convolution Theorem

CpE 584 Al-Omari


Rotation

Rotating a 2D function, rotates its Fourier


Transform

f 2 = rot ( f1 )
if

Then,
F2 = rot (F1 )
i.e., the Fourier Transform is Rotation Invariant

CpE 584 Al-Omari


Rotation

The point (u , v) in the frequency domain


corresponds to the basis function with
frequency u in x and frequency v in y.
OR
The point (u , v) in the frequency domain
corresponds to the basis function with
frequency |(u , v)| in the direction (u , v).

CpE 584 Al-Omari


Transforms of Separable Functions

If,
f ( x, y ) = f ( x ) f ( y )
Then, the function is called separable
Its transform is also separable

F (u ,v ) = F (u )F (v )

CpE 584 Al-Omari


Linear Separability of Fourier
Transform

The 2-D Fourier Transform is linearly


separable
The Fourier Transform of a two-dimensional
image is the Fourier Transform of the rows
followed by the Fourier Transforms of the
resulting columns (or vice versa).

CpE 584 Al-Omari


Linear Separability of Fourier
Transform

N 1 M 1
F [u ,v ] = f [x, y ]e j2 (ux / N + vy / M )
1
MN
x =0 y =0

N 1 M 1
F [u ,v ] = f [x, y ]e j2 (ux / N ) e j2 ( vy / M )
1
MN

x =0 y =0

N 1 1 M 1 j2 (ux / N )
F [u ,v ] = f [x, y ]e
1

x =0 M
j2 ( v y / M )
e
N y =0
CpE 584 Al-Omari
Computational Complexity

The order of computations for the 2D Fourier


Transform:
Convolution O(N4)
DFT O(N4)
DFT using separability O(N3)
FFT using separability O(N2 logN)

CpE 584 Al-Omari


Convolution Using FFT

Convolution Theorem says:


f g = F -1
(F ( f )F (g ))
Can do either,
Convolution
FFT, multiplication, and inverse FFT
Computational breakeven point: about 7 x 7
kernel

CpE 584 Al-Omari


Boundaries

Unanswerable question: what do you do


when the shifted convolution kernel extends
outside the bounds of the image?
Simple pad with zeroes
Better pad with the average grey value of
the image
Or pad by replicating the last element(s)

CpE 584 Al-Omari


Spatial Frequencies

If the image makes gradual transitions, it only


requires low-frequency sinusoids.
If the image makes rapid transitions, it requires
high-frequency sinusoids.
Places with low spatial frequency content:
smooth regions
Places with high spatial frequency content:
edges, texture

CpE 584 Al-Omari


Example: Grey Level Image

CpE 584 Al-Omari


Example: Image Fourier Transform
(Magnitude)

CpE 584 Al-Omari


Example: Image Fourier Transform
(Mag.) Quadrants Reshuffled

CpE 584 Al-Omari


Example: Image Fourier Transform
(Phase) Quadrants Reshuffled

CpE 584 Al-Omari


Example 2

CpE 584 Al-Omari


Example 2

CpE 584 Al-Omari

You might also like