IMAGE PROCESSING

Filtering in frequency domain

Le Thanh Ha, Ph.D

Assoc. Prof. at University of Engineering and Technology,
Vietnam National University
[email protected]; [email protected]; 0983 692 592
 - Discrete Fourier transform
- Filtering in frequency domain
- Aliasing
- Hybrid image
- Common Frequency filters

THINKING IN FREQUENCY

12/21/2024 Le Thanh Ha 2
 Frequency Spectra
• example : g(t) = sin(2πf t) + (1/3)sin(2π(3f) t)

= +

Slides: Efros
Frequency Spectra
Frequency Spectra

= +

=
Frequency Spectra

= +

=
Frequency Spectra

= +

=
Frequency Spectra

= +

=
Frequency Spectra

= +

=
Frequency Spectra


1
= A sin(2 kt )
k 1 k
 1D Discrete Fourier transform

• Complex series: Xk, xn

N1 2 i
 kn
X k  xn e N
k 0,..., N  1
n 0
N1 2 i
1 kn
xn 
N
X
k 0
k e N
n 0,..., N  1

12/21/2024 Le Thanh Ha 11
 2D Discrete Fourier transform
• Discrete domain:

M N
F ( wx , wy )   f ( x, y )e
 i ( wx x  wy y )

x 1 y 1
M N
1
  F (wx , wy )e
i ( wx x  wy y )
f ( x, y ) 
MN x 1 y 1

12/21/2024 Le Thanh Ha 12
 2D Fourier transform

• f – image, matrix of real numbers

• F – frequency image, matrix of complex
numbers.

MAGNITUDE(F) = SQRT(real(F)2 +imag(F)2) – Energy of

frequency
PHASE(F) = ATAN(imag(F)/real(F) )

Which one is more informative, magnitude or phase?

12/21/2024 Le Thanh Ha 13
 Fourier analysis in images

Intensity Image

Fourier Image

http://sharp.bu.edu/~slehar/fourier/fourier.html#filtering
Signals can be composed

+ =

http://sharp.bu.edu/~slehar/fourier/fourier.html#filtering
More: http://www.cs.unm.edu/~brayer/vision/fourier.html
12/21/2024 Le Thanh Ha 16
 2D Fourier transform

Left: shows the vertical

and horizontal
frequencies as the
edges of bricks.

Right: shows that the

edges generate high
frequencies with
high energy.
12/21/2024 Le Thanh Ha 17
 2D Fourier transform

FT of each character
symbol is different
from others,
especially at the
low frequencies.

12/21/2024 Le Thanh Ha 18
 2D Fourier transform

• 2 real images

• FTs show that:

– The high
frequencies of the
left image is less
than those of the
right image.

12/21/2024 Le Thanh Ha 19
 2D Fourier transform

• Ideal low-pass filter:

– Completely remove
the high frequencies
of the left image.
– The right image is
obtained by
applying the inverse
FT (IFT)

12/21/2024 Le Thanh Ha 20
 2D Fourier transform

• High pass filter

– Remove allmost all
low frequencies
– The resulted image is
too dard because high
energy is condensed in
low frequencies which
are removed.
– There edges left (high
frequencies)
12/21/2024 Le Thanh Ha 21
 2D Fourier transform - IP

• The left image is

added with horizontal
wave
• There are very bright
dots in frequency
domain.
• The image looks good
after remove the noise
frequency.
12/21/2024 Le Thanh Ha 22
 Demonstration using MATLAB
• Operator order:
– Read an Image
– Transform using fft2
– Shift frequency coefs to center using fftshift
– Create filters (ideal, gaussian, …)
– Filter in frequency domain
– Shift frequency coefs back using ifftshift
– Invert transform using ifft2

12/21/2024 Le Thanh Ha 23
 Practice question
Match the spatial domain image to the Fourier magnitude image

1 2 3 4 5

A C E
D
 Sampling
Why does a lower resolution image still make
sense to us? What do we lose?

Image: http://www.flickr.com/photos/igorms/136916757/
 Subsampling by a factor of 2

Throw away every other row and column

to create a 1/2 size image
 Aliasing problem
• 1D example (sinewave):

Source: S. Marschner
 Aliasing problem
• 1D example (sinewave):

Source: S. Marschner
 Aliasing problem
• Sub-sampling may be dangerous….
• Characteristic errors may appear:
– “Wagon wheels rolling the wrong way in
movies”
– “Checkerboards disintegrate in ray tracing”
– “Striped shirts look funny on color television”

Source: D. Forsyth
Aliasing in video

Slide by Steve Seitz

Aliasing in graphics

Source: A. Efros
Sampling and aliasing
Nyquist-Shannon Sampling Theorem
• When sampling a signal at discrete intervals, the
sampling frequency must be  2  fmax
• fmax = max frequency of the input signal
• This will allows to reconstruct the original
perfectly from the sampled version

v v v good

bad
 Anti-aliasing

Solutions:
• Sample more often

• Get rid of all frequencies that are greater than half the new
sampling frequency
– Will lose information
– But it’s better than aliasing
– Apply a smoothing filter
Algorithm for downsampling by factor of 2
1. Start with image(h, w)
2. Apply low-pass filter
im_blur = imfilter(image, fspecial(‘gaussian’, 7, 1))
3. Sample every other pixel
im_small = im_blur(1:2:end, 1:2:end);
 With and without anti-aliasing

12/21/2024 Le Thanh Ha 36
Why does a lower resolution image still make
sense to us? What do we lose?

Image: http://www.flickr.com/photos/igorms/136916757/
 What do you see?

12/21/2024 Le Thanh Ha 38
Why do we get different, distance-dependent interpretations of
hybrid images?

Come closer

Go further
 Hybrid Images

• A. Oliva, A. Torralba, P.G. Schyns,

“Hybrid Images,” SIGGRAPH 2006
•
Clues from Human Perception
Early processing in humans filters for various orientations and scales of
frequency
• Perceptual cues in the mid-high frequencies dominate perception
• When we see an image from far away, we are effectively subsampling it

Early Visual Processing: Multi-scale edge and blob filters

 Scaling

12/21/2024 Le Thanh Ha 42
 Subband filter

50 50 50

100 100 100

150 150 150

200 200 200

250 250 250

300 300 300

50 100 150 200 250 300 50 100 150 200 250 300 50 100 150 200 250 300

12/21/2024 Le Thanh Ha 43
 Ideal Lowpass filter

• Cut off high frequencies specified by a distance d 0.

– Cannot be realized by electronic component  not
practical
– Causes ringing effect  How to remove

12/21/2024 Le Thanh Ha 44
 Ideal Lowpass filter

12/21/2024 Le Thanh Ha 45
 Ideal Lowpass filter

12/21/2024 Le Thanh Ha 46
 Butterworth Filter

1
H (u , v) 
1   D(u , v) / D0 
n

12/21/2024 Le Thanh Ha 47
 Butterworth Filter

12/21/2024 Le Thanh Ha 48
 Butterworth Filter

12/21/2024 Le Thanh Ha 49
 Gaussian Filter

12/21/2024 Le Thanh Ha 50
 Gaussian Filter

12/21/2024 Le Thanh Ha 51
12/21/2024 Le Thanh Ha 52
 Demonstration using MATLAB
• Operator order:
– Read an Image
– Transform using fft2
– Shift frequency coefs to center using fftshift
– Create filters (ideal, butterworth, …)
– Filter in frequency domain
– Shift frequency coefs back using ifftshift
– Invert transform using ifft2

12/21/2024 Le Thanh Ha 53