Computer Vision

Spring 2012 15-385,-685

Instructor: S. Narasimhan

Wean Hall 5409

T-R 10:30am – 11:50am

Frequency domain analysis and Fourier Transform

Lecture #4

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:

• 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?

xAsin(

Fourier Transform

• We want to understand the frequency of our signal. So, let’s reparametrize the signal by instead of x:

xAsin(

f(x) F()Fourier Transform

F() f(x)Inverse Fourier Transform

• For every from 0 to inf, F() holds the amplitude A and phase of the corresponding sine

– How can F hold both? Complex number trick!

)()()( iIRF

22 )()( IRA )(

)(tan 1

R

I

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

= 1

1sin(2 )

k

A ktk

Frequency Spectra

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

Arbitrary function Single Analytic Expression

Spatial Domain (x) Frequency Domain (u)

Represent the signal as an infinite weighted sum of an infinite number of sinusoids

dxexfuF uxi 2

(Frequency Spectrum F(u))

1sincos ikikeikNote:

Inverse Fourier Transform (IFT)

dxeuFxf uxi 2

• Also, defined as:

dxexfuF iux

1sincos ikikeikNote:

• Inverse Fourier Transform (IFT)

dxeuFxf iux

21

Fourier Transform

Fourier Transform Pairs (I)

angular frequency ( )iuxeNote that these are derived using

angular frequency ( )iuxeNote that these are derived using

Fourier Transform Pairs (I)

Fourier Transform and Convolution

hfg

dxexguG uxi 2

dxdexhf uxi 2

dxexhdef xuiui 22

'' '22 dxexhdef uxiui

Let

Then

uHuF

Convolution in spatial domain

Multiplication in frequency domain

Fourier Transform and Convolution

hfg FHG fhg HFG

Spatial Domain (x) Frequency Domain (u)

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

g f h

G F H

FT FTIFT

Properties of Fourier Transform

Spatial Domain (x) Frequency Domain (u)

Linearity xgcxfc 21 uGcuFc 21

Scaling axf

a

uF

a

1

Shifting 0xxf uFe uxi 02

Symmetry xF uf

Conjugation xf uF

Convolution xgxf uGuF

Differentiation n

n

dx

xfd uFui n2

frequency ( )uxie 2Note that these are derived using

Properties of Fourier Transform

Example use: Smoothing/Blurring

• We want a smoothed function of f(x)

xhxfxg

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

• Then

222

2

1exp uuH

uHuFuG

2

1

u

uH

2

2

2

1exp

2

1

x

xh

• Let us use a Gaussian kernel

xh

x

Does not look anything like what we have seen

Magnitude of the FT

Image Processing in the Fourier Domain

Image Processing in the Fourier Domain

Does not look anything like what we have seen

Magnitude of the FT

Convolution is Multiplication in Fourier Domain

*

f(x,y)

h(x,y)

g(x,y)

|F(sx,sy)|

|H(sx,sy)|

|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