Professional Documents
Culture Documents
Spectral
Techniques
Feb 11, 12, 2019
2
Spatial a
and
Frequency domain time domain frequency domain
b
d=a+b+c
d
time domain frequency domain
c
time domain frequency domain
• Solution?
– To remove certain frequencies, set their corresponding F(u)
coefficients to zero! See how it can be done…
11
Approximating a
square wave as
the sum of sine
waves
18
Phase (Recall)
• In general, the function f(x-d) is the original
function f(x) shifted to the right by a distance „d‟
• Shifting a cosine function along the x-axis by a
distance :
Phase
• Cosine and sine functions are “orthogonal” in a sense
that they can be used to create new “sinusoidal”
functions with arbitrary frequency, phase, and
amplitude
Fourier Integral
• For non-periodic functions we can get similar
results by letting 𝑇 ∞
• Similar ideas yield Fourier Integral
– Integration of densely packed sines and cosines
Fourier Integral
reconstruct the original
function from its spectrum
Fourier Transform
• Fourier Transform: Transition of function 𝑔(𝑥) to
its Fourier spectrum 𝐺(𝜔)
g(x) GRe(ω)
g(x) GRe(ω)
Cos 3x
g(x) GIm(ω)
Cos 5x
g(x) GIm(ω)
Sin 3x
Sin 5x
31
g(x) GRe(ω)
σ=1
σ=3
g(x) GRe(ω)
b=1
2𝑏 sin(𝑏𝜔)
𝐺 𝜔 = b=2
2𝜋(𝑏𝜔)
32
• Symmetry
• Linearity
• Similarity/Scaling
• Shift property
• Convolution property
Convolution in spatial domain
Multiplication (pointwise) in frequency domain
33
Sampling
• A continuous function 𝑓(𝑥) is discretized as:
{ f ( x0 ), f ( x0 x), f ( x0 2x),..., f ( x0 (M 1)x)}
Impulse Function
∞ if 𝑥 = 0
𝛿 𝑥 =
0 if 𝑥 ≠ 0
The continuous
signal g(x) is
sampled at
position x0 = 3
sum of pulse
shifted sequence or
impulses pulse train
36
Discrete Signals
• If a continuous signal 𝑔(𝑥) is
sampled at regular intervals 𝜏, then
the corresponding Fourier spectrum
becomes periodic with a period of
length 𝜔𝑠 = 2𝜋 𝜏
• Sampling in signal space leads to
periodicity in frequency space and
vice versa
• Image is a 2D function!
– For discrete functions we need only finite
number of sequences
– Ex: consider the discrete sequence
1, 1, 1, 1, -1, -1, -1, -1, -1
It is discrete approximation of square wave
References
• Digital Image Processing - An
Algorithmic Introduction using Java by
Dr. Wilhelm Burger and Dr. Mark J. Burge
• R. C. Gonzalez and R. E. Woods, Digital
Image Processing, Third Edition,
Pearson, 2012.
• Computer Vision course by Dr. S.
Narasimhan