ECE 472/572 - Digital Image

Processing

Lecture 5 - Image Enhancement -

Frequency Domain Filters
09/13/11
2
Why FT? – 1

3
Why FT? – 2

4
Another example

5
Fourier series

 Fourier series (SF) can represent any function

over a finite interval TF
 Outside TF, SF repeats periodically with period
TF.

6
Fourier series (cont’)

 TF is the interval of signal s(t) over which the Fourier

series represents
 fF = 1/TF is the fundamental frequency of the Fourier
series representation
 n is called the “harmonic number”
– E.g., 2fF is the second harmonic of the fundamental frequency fF.
 The Fourier series representation is always periodic and
is linear combinations of sinusoids at fF and its
harmonics.

7
 Fourier transform

Describe the frequency distribution

cn

(0,0) u

nfF v

8
9
 1-D Fourier transform

 Fourier transform:
DFT
 Inverse FT:

 Complex form

 Fourier spectrum
 Power spectrum
(spectral density)
 Phase angle
10
2-D Fourier transform

CFT

DFT

11
Understanding and implementing
Fourier transform

(0,0) y
(0,0) v

x u
12
f(x,y) |F(u,v)|
13
 Understanding and implementing
Fourier transform

According to “translation”

y
(0,0) y (0,0)

0 -255 v

-0 255 (0,0)

x x
u
f(x,y) f(x,y)(-1)x+y |F(u-M/2,v-N/2)|

14
Understanding and implementing
Fourier transform

15
 Block diagram of FTIFT

f(x,y)(-1)x+y g(x,y)(-1)x+y

16
17
Impulse transforms

18
Typical transforms

 Gaussian hump 
Gaussian hump

 Rectangular (square
aperture) sinc

 Pillbox (circular
aperture) jinc
19
Typical transforms

Gaussian ridge

Line impulse

20
Reference

All figures scanned from R. N. Bracewell’s

“Two-Dimensional Imaging,” Prentice
Hall, 1995.

21
2D FT pairs

22
Important properties of FT

 Linearity (distributivity && scaling)

 Separability
 Translation
 Periodicity
 Conjugate symmetry
 Rotation
 Convolution
 Correlation
 Sampling

23
 Linearity

x1(t) y1(t)
Linear
x2(t) y2(t)
System

a*x1(t) + b*x2(t) a*y1(t) + b*y2(t)

FT is a linear image processing method

24
Separability

Row Column
transform transform
f(x, y) F(x, v) F(u, v)

25
Translation

26
Periodicity and Conjugate
symmetry
 Periodicity

 Conjugate symmetry

27
Example

28
Rotation

29
Averaging

30
Convolution

 Continuous and discrete convolution

 The convolution theorem

 Practically, computing the discrete convolution in the frequency

domain often is more efficient than doing it in the spatial domain
directly
31
Correlation

 Continuous and discrete correlation

 The correlation theorem

32
Correlation (cont’)

 Autocorrelation vs. cross correlation

 Autocorrelation theorem

 Application: template or prototype matching

33
Practical issues – Implement
convolution in frequency domain

 In spatial domain

 In frequency domain
– f*g  F(f)G(g)
– Phase? Mag?
– How to pad?

34
35
36
37
38
 Difference image
from convolution
in the spatial
domain

Convolution in
the frequency
domain

No padding With padding

Conv. spatially 39
Different enhancement
approaches
Lowpass filter
Highpass filter
Homomorphic filter

40
Lowpass filtering

 Ideal filter
– D(u, v): distance from point (u, v) to the origin
– cutoff frequency (D0)
– nonphysical
– radially symmetric about the origin
 Butterworth filter

 Gaussian lowpass filter

41
42
Power ratio
99.9699.6599.0497.84

43
Highpass filter

Ideal filter

Butterworth filter

Gaussian highpass filter

44
45
46
Example

47
The Laplacian in the frequency domain

48
UM in the frequency domain

49
Homomorphic filtering

A simple image model

– f(x,y): the intensity is called the gray level for
monochrome image
– f(x, y) = i(x, y).r(x, y)
– 0 < i(x, y) < inf, the illumination
– 0< r(x, y) < 1, the reflectance

50
Homomorphic filter (cont’)

51
Homomorphic filter (cont’)

 The illumination component

– Slow spatial variations
– Low frequency
 The reflectance component
– Vary abruptly, particularly at the junctions of
dissimilar objects
– High frequency
 Homomorphic filters
– Affect low and high frequencies differently
– Compress the low frequency dynamic range
– Enhance the contrast in high frequency 52
Homomorphic Filter (cont’)

53
54
Homomorphic filter - example

55
 Point processing  Mask processing • Frequency domain
filters
(spatial filters)
 Simple gray level  Smoothing filters (blur • Smoothing filters (blur
transformations details) details)
• Ideal lowpass filter
– Image negatives – Average, weighted • Butterworth lowpass
– Log transformations average • Gaussian lowpass
– Power-law – Order statistics (e.g. • Sharpening filters (highlight
transformations median) details)
– Contrast stretching – Unsharp masking
– High-boost filters
– Gray-level slicing  Sharpening filters – Derivative filters - The
– Bit-plane slicing (highlight details) Laplacian
– Ideal highpass filter
 Histogram processing – Unsharp masking
– Butterworth highpass filter
– Histogram – High-boost filters – Gaussian highpass filter
equalization – Derivative filters • Homomorphic filtering
– *Histogram matching • The Laplacian
(specification) • The Gradient
 Arithmetic/logic
operations
– Image averaging
56
FFT and IFFT

57