Digital Image Processing

& Pattern Recognition

Lecture#5
Image Enhancement in the Frequency
Domain

Prepared by: Dr. Gawed Nagi

Contents
This lecture will cover:
◼ Introduction to the Fourier Transform and
the Frequency Domain
◼ Smoothing Frequency-Domain Filters
◼ Sharpening Frequency-Domain filters

2
What is the Fourier Transform
Jean Baptiste Joseph Fourier (1768~1830)
◼ French mathematician

3
What is the Fourier Transform
 the Fourier Transform dates back more than
200 years.
 In 1807 he found that periodic functions could
be written as a summation of sines and
cosines.
 Fourier published his findings as part of The
Analytical Theory of Heat in 1822.
 Two types of Fourier Transforms are commonly
used today in computer based applications:
◼ the Discrete Fourier Transform (DFT),
◼ the Fast Fourier Transform (FFT)
4
What is the Fourier Transform

Any function that periodically repeats itself can be

expressed as a sum of sines and cosines of different
frequencies each multiplied by a different coefficient –
a Fourier Series 5
What is the Fourier Transform

6
What is the Fourier Transform
 Prism: device that separates light into various
color components each depending on its
wavelength (frequency) content.
 Fourier Transform: mathematical prism that
separates a function into various components
based on frequency contents.

7
What is the Fourier Transform

8
Fourier Transform
 We will be dealing only with functions
(images) of finite duration so we will be
interested only in Fourier Transform

9
Applications of Fourier Transforms
 1-D Fourier transforms are used in Signal
Processing
 2-D Fourier transforms are used in Image
Processing
 3-D Fourier transforms are used in Computer
Vision
 Applications of Fourier transforms in Image
processing:
◼ Image enhancement,
◼ Image restoration,
◼ Image encoding / decoding,
◼ Image description 10
Discrete Fourier Transforms (DFT)
1-D DFT for M samples is given as:

for u=1,2, ……., m-1

The Inverse Fourier transform in 1-D is given as:

for x=1,2, ……., m-1

11
Discrete Fourier Transforms (DFT)

12
2-D DFT

13
Discrete Fourier Transforms (DFT)

14
Discrete Fourier Transforms (DFT)
Displaying the FT of an Image
 Shifting the center to place the largest
energy in the middle and be able to
observe the behavior

15
Discrete Fourier Transforms (DFT)

16
Discrete Fourier Transforms (DFT)
Displaying the FT of an Image
 It is common practice to scale the FT to be
able to display all values in the range.

17
Discrete Fourier Transforms (DFT)
Example

18
Properties of Fourier Transform
Translation & Rotation

19
Properties of Fourier Transform
Remarks
 The DC component (lowest frequency) has
the largest magnitude.

20
Low Frequency & High Frequency
 Low Frequency: slowly varying components
on image such as backgrounds and steady
areas.
 High Frequency: faster gray level changes
such as noise and edges of objects.

21
Frequency Domain Filters
Why Frequency Domain Filters
 It makes more sense to filter in the special
domain using small filter masks, but it is
more computationally efficient to do the
filtering in frequency domain.
 Filtering is more controllable in the frequency
domain.

22
Frequency Domain Filters
Convolution Theorem
 Convolution in the spatial domain corresponds
to multiplication in the frequency domain, and
vice versa.
f(x,y)*h(x,y) F(u,v)H(u,v)

 This theorem forms the basic step of

filtering the frequency domain.

23
 Frequency Domain Filters

24
Lowpass & Highpass Filters

Original Image

25
Frequency Domain Filters

26
Ideal Lowpass Filter (ILPF)
 Cuts off all high frequency components
that are a specified distance D0 from the
origin of the transform

27
Ideal Lowpass Filter (ILPF)
 The transform function is given by:

 Changing the distance changes the effect

of the filter
 Point of transition between H(u,v) = 1 and
H(u,v) = 0 is called cutoff frequency
28
Ideal Lowpass Filter (ILPF)

29
Ideal Lowpass Filter (ILPF)

30
Ringing Effect

31
Ideal Lowpass Filter (ILPF)

32
Ideal Lowpass Filter (ILPF)

33
Butterworth Lowpass Filter (BLPF)
 Does not have the sharp discontinuity of
the ideal filter.
 Represents the transition between two
extremes.

34
Butterworth Lowpass Filter (BLPF)
 The transfer function is given by:

 The order n of the filter controls the

“degree” of smoothness.
◼ Higher order -> sharper cutoff ~~ ideal.
◼ Lower order -> smoother cutoff ~~ Gaussian.

35
Butterworth Lowpass Filter (BLPF)

36
Butterworth Lowpass Filter (BLPF)

37
Butterworth Lowpass Filter (BLPF)
Remarks
 No ringing effect for small order
 Smoother transitions is blurring with
respect to increasing cutoff.

38
Gaussian Lowpass Filter (GLPF)
 The transfer function is given by:

39
Gaussian Lowpass Filter (GLPF)

40
Gaussian Lowpass Filter (GLPF)
 Optimal Filter: The Gaussian

41
Gaussian Lowpass Filter (GLPF)

460

42
Gaussian Lowpass Filter (GLPF)
 Example: for the broken char. The human
visual system can fill these gaps, but the
machine is not.
 A lowpass gaussian filter is used to connect
broken text (blurring)

43
Gaussian Lowpass Filter (GLPF)
 Example: different lowpass gaussian filters
used to remove blemishes in a photograph.

44
Lowpass Filters Effects Compared

45
Highpass Filters
 The filter cut off (set to zero) all low
frequency components that are a specified
distance D0 from the original of the
transform
 Highpass filter can be obtained from lowpass
filter by:

46
Highpass Filters

47
Ideal Highpass Filter (IHPF)
 Transfer function is given by:

48
Ideal Highpass Filter (IHPF)

49
Butterworth Highpass Filter (BHPF)
 Transfer function is given by:

50
Gaussian Highpass Filter (GHPF)
 Transfer function is given by:

51
Gaussian Highpass Filter (GHPF)

52
Highpass Filters Effects Compared
 Ringing Effects

53
Highpass Filters Effects Compared

54
Denoising in Frequency Domain
 The type of noise determines the domain and
method of denoising the image.
 Sources of noise in digital images arise
during image acquisition and transmission
◼ Imaging sensors can be affected by ambient
conditions -> random noise.
◼ Interference can be added to an image during
transmission -> periodic noise.

55
Denoising in Frequency Domain
 Periodic Noise
◼ From electrical or
electromechanical
interference during
acquisition
◼ Spatially
dependent
◼ Can be reduced
significantly by
frequency domain
filtering

56
Denoising in Frequency Domain
Selective Filters: Bandreject Filters

57
Denoising in Frequency Domain
Selective Filters: Bandreject Filters

Goodness Measures: Mean Square Error

58
Q&A

59