Computer Vision

How to Represent Signals?

• Option 1: Taylor series represents any function using
polynomials.

• Polynomials are not the best - unstable and not very

physically meaningful.

• Easier to talk about “signals” in terms of its “frequencies”

(how fast/often signals change, etc).
 Jean Baptiste Joseph Fourier (1768-1830)
• Had crazy idea (1807):
• Any periodic function
can be rewritten as a
weighted sum of Sines and
Cosines of different
frequencies.
• Don’t believe it?
– Neither did Lagrange,
Laplace, Poisson and
other big wigs
– Not translated into
English until 1878!
• But it’s true!
– called Fourier Series
– Possibly the greatest tool
used in Engineering
 A Sum of Sinusoids
• Our building block:

Asin(x   
• Add enough of them to
get any signal f(x) you
want!

• How many degrees of

freedom?

• What does each control?

• Which one encodes the

coarse vs. fine structure of
the signal?
 Fourier Transform
• We want to understand the frequency of our signal. So, let’s
reparametrize the signal by  instead of x:

Fourier
f(x) F()
Transform

• For every  from 0 to inf, F() holds the amplitude A and phase of the
corresponding sine
Asin(x   
– How can F hold both? Complex number trick!

F ( )  R( )  iI ( )
I ( )
A   R( )  I ( )
2 2
  tan 1

R( )
Inverse Fourier
F() f(x)
Transform
 Time and Frequency

• example : g(t) = sin(2pi f t) + (1/3)sin(2pi (3f) t)

 Time and Frequency

• example : g(t) = sin(2pi f t) + (1/3)sin(2pi (3f) t)

= +
 Frequency Spectra

• example : g(t) = sin(2pi f t) + (1/3)sin(2pi (3f) t)

= +
 Frequency Spectra
• Usually, frequency is more interesting than the phase
Frequency Spectra

= +

=
Frequency Spectra

= +

=
Frequency Spectra

= +

=
Frequency Spectra

= +

=
Frequency Spectra

= +

=
Frequency Spectra


1
= A sin(2 kt )
k 1 k
Frequency Spectra
 FT: Just a change of basis

M * f(x) = F()

* =

.
.
.
IFT: Just a change of basis

M-1 * F() = f(x)

* =

.
.
.
 Fourier Transform – more formally
Represent the signal as an infinite weighted sum
of an infinite number of sinusoids

F  u    f  xe i 2ux
dx


Note: e
ik
 cos k  i sin k i  1

Arbitrary function Single Analytic Expression

Spatial Domain (x) Frequency Domain (u)
(Frequency Spectrum F(u))

Inverse Fourier Transform (IFT)


f  x   F  u e i 2ux
dx

 Fourier Transform

• Also, defined as:


F  u    f  xe iux
dx


Note: e
ik
 cos k  i sin k i  1

• Inverse Fourier Transform (IFT)

1 
f  x   F  u  e iux dx
2 
Fourier Transform Pairs (I)

Note that these are derived using

angular frequency ( e  iux )
Fourier Transform Pairs (I)

Note that these are derived using

angular frequency ( e  iux )
 Fourier Transform and Convolution
Let g  f h

Then G  u    g  x  e i 2ux dx
 
  f   h x    e i 2ux ddx
 

  f   e  
 
  i 2u
d h x    e i 2u  x   dx
 





 f   e  i 2u
d    h  x ' e



i 2ux '

dx'

 F u H u

Convolution in spatial domain

 Multiplication in frequency domain
Fourier Transform and Convolution

Spatial Domain (x) Frequency Domain (u)

g  f h G  FH
g  fh G  F H

So, we can find g(x) by Fourier transform

g  f  h
IFT FT FT

G  F  H
 Properties of Fourier Transform
Spatial Domain (x) Frequency Domain (u)

Linearity c1 f  x   c2 g  x  c1 F  u   c2G  u 
1 u
Scaling f  ax  F 
a a
Shifting f  x  x0  e  i 2ux0 F  u 
Symmetry F  x f   u
Conjugation f   x F   u
Convolution f  x  g  x F  u G u 
d n f  x
Differentiation  i 2u  F  u 
n

dx n
Note that these are derived using
frequency ( e  i 2ux )
Properties of Fourier Transform
 Example use: Smoothing/Blurring
• We want a smoothed function of f(x)

g  x   f  x   h x 

• Let us use a Gaussian kernel h x 

1  1 x2 
h x   exp  2 
2   2  
x
• Then H u
 1 2 2
1
H  u   exp   2u    2
 2 
G u   F  u  H  u  u

H(u) attenuates high frequencies in F(u) (Low-pass Filter)!

Image Processing in the Fourier Domain
Magnitude of the FT

Does not look anything like what we have seen

Image Processing in the Fourier Domain

Magnitude of the FT

Does not look anything like what we have seen

 Convolution is Multiplication in Fourier Domain

f(x,y) |F(sx,sy)|

h(x,y) |H(sx,sy)|

g(x,y) |G(sx,sy)|
 Low-pass Filtering

Let the low frequencies pass and eliminating the high frequencies.

Generates image with overall

shading, but not much detail
 High-pass Filtering

Lets through the high frequencies (the detail), but eliminates the low
frequencies (the overall shape). It acts like an edge enhancer.
Boosting High Frequencies
Most information at low frequencies!
Fun with Fourier Spectra
 Next Class

• Image resampling and image pyramids

• Horn, Chapter 6