Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals –...

26
Signals and Systems Ho Kyung Kim Pusan National University Medical Physics Prince & Links 2 Signals Mathematical functions capable of modeling a variety of physical processes Continuous and discrete signals , , െ∞,∞ function or signal at a point , image at a pixel , Systems Response to signals by producing new signals (output, e.g., images) Continuous‐to‐continuous & continuous‐to‐discrete (e.g., ADC by sampling) systems , ሺ, ሻ output signal by applying the transformation on the entire signal ሺ, ሻ The challenge in medical imaging is to make the output signal (as coming from the imaging system) a faithful representation of the input signal (as coming from the patient) 2

Transcript of Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals –...

Page 1: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

Signals and Systems

Ho Kyung Kim

Pusan National University

Medical PhysicsPrince & Links 2

• Signals

– Mathematical functions capable of modeling a variety of physical processes

– Continuous and discrete signals

• 𝑓 𝑥, 𝑦 , ∞ 𝑥, 𝑦 ∞– function or signal at a point 𝑥, 𝑦– image at a pixel 𝑥, 𝑦

• Systems

– Response to signals by producing new signals (output, e.g., images)

– Continuous‐to‐continuous & continuous‐to‐discrete (e.g., ADC by sampling) systems

• 𝑔 𝑥, 𝑦 𝒮 𝑓 𝑥, 𝑦– output signal by applying the transformation 𝑆 on the entire signal 𝑓 𝑥, 𝑦

• The challenge in medical imaging is to make the output signal (as coming from the imaging system) a faithful representation of the input signal (as coming from the patient)

2

Page 2: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

3

Functional plot & image display for signal visualizationRepresentation of a 3D object as a 2D image

Response of a continuous system to a point impulse

Point impulse

• Useful to model a "point source" when characterizing the imaging system resolution

• 1D point impulse

– 𝛿 𝑥 0, 𝑥 0• aka delta function, Dirac function, impulse function

• Not a function

– 𝑓 𝑥 𝛿 𝑥 d𝑥 𝑓 0

• 2D point impulse

– 𝛿 𝑥, 𝑦 0, 𝑥, 𝑦 0,0– 𝛿 𝑥, 𝑦 𝛿 𝑥 𝛿 𝑦 separable signal

– 𝑓 𝑥, 𝑦 𝛿 𝑥, 𝑦 d𝑥d𝑦 𝑓 0,0

– 𝑓 𝜉, 𝜂 𝑓 𝑥, 𝑦 𝛿 𝑥 𝜉, 𝑦 𝜂 d𝑥d𝑦 shifting property

• "picking off" the value of a function at a point

– 𝛿 𝑎𝑥, 𝑏𝑦 𝛿 𝑥, 𝑦 scaling property

– 𝛿 𝑥, 𝑦 𝛿 𝑥, 𝑦 even function

4

Page 3: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

5

Point & line impulses

Example

• What is the nature of the function 𝑔 𝑥, 𝑦 𝑓 𝑥, 𝑦 𝛿 𝑥 𝜉, 𝑦 𝜂 ?

6

Page 4: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

Line impulse

• A set of points describing a line whose unit normal is oriented at an angle 𝜃 relative to the x‐axis and is at distance ℓ from the origin in the direction of the unit normal:

– 𝐿 ℓ, 𝜃 𝑥, 𝑦 |𝑥 cos 𝜃 𝑦 sin 𝜃 ℓ

• Line impulse

– 𝛿ℓ 𝑥, 𝑦 𝛿 𝑥 cos 𝜃 𝑦 sin 𝜃 ℓ)– 1‐D impulse function “spread out” on the line 𝐿 ℓ, 𝜃

7

Quantum & digital images

-10 -5 0 5 10

x (mm)

10-6

10-4

10-2

100

a = 1 m a = 10 m a = 50 m a = 100 m

0 5 10 15

f (mm-1)

0.0

0.2

0.4

0.6

0.8

1.0 a = 1 m a = 10 m a = 50 m a = 100 m

8

∆𝑎 10 𝜇𝑚 ∆𝑎 50 𝜇𝑚

Thanks to Junwoo for preparing this slide

-2 -1 0 1 2

x (mm)

Page 5: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

Comb & sampling functions

• Sampling

– "Picking off" values on a grid or matrix of points

• Comb function (aka the shah function)

– comb 𝑥, 𝑦 ∑ ∑ 𝛿 𝑥 𝑚, 𝑦 𝑛

• Sampling function

– 𝛿 𝑥, 𝑦; ∆𝑥, ∆𝑦 ∑ ∑ 𝛿 𝑥 𝑚∆𝑥, 𝑦 𝑛∆𝑦• A sequence of point impulses located at points  𝑚∆𝑥, 𝑛∆𝑦 , ∞ 𝑚, 𝑛 ∞

– 𝛿 𝑥, 𝑦; ∆𝑥, ∆𝑦∆ ∆

comb∆

,∆

from the scaling property

9

Rect & sinc functions

• Rect function

– rect 𝑥, 𝑦1, for 𝑥 & 𝑦

0, for 𝑥 & 𝑦

– rect 𝑥, 𝑦 rect 𝑥 rect 𝑦 separable signal

• Finite energy signal, with unit total energy, that models signal concentration over a unit square centered around point  0,0

– 𝑓 𝑥, 𝑦 rect ,

• To select part of signal 𝑓 𝑥, 𝑦 centered at point  𝜉, 𝜂 of the plane with width 𝑋 and height 𝑌, and set the rest to zero

10

Page 6: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

• Sinc function

– sinc 𝑥, 𝑦1, for 𝑥 𝑦 0

, otherwise

– sinc 𝑥, 𝑦 sinc 𝑥 sinc 𝑦 separable signal

• Again, finite energy signal with unit total energy

11

Main lobe

Side lobes

Exponential & sinusoidal signals

• Complex exponential signals

– 𝑒 𝑥, 𝑦 𝑒• 𝑢 & 𝑣 in units of the inverse of the units of 𝑥 & 𝑦, hence mm‐1, are referred to as 

fundamental frequencies

– 𝑒 𝑥, 𝑦 cos 2𝜋 𝑢 𝑥 𝑣 𝑦 𝑗 sin 2𝜋 𝑢 𝑥 𝑣 𝑦 𝑐 𝑥, 𝑦 𝑗𝑠 𝑥, 𝑦

• Sinusoidal signals

– 𝑠 𝑥, 𝑦 sin 2𝜋 𝑢 𝑥 𝑣 𝑦 𝑒 𝑒

– 𝑐 𝑥, 𝑦 cos 2𝜋 𝑢 𝑥 𝑣 𝑦 𝑒 𝑒

• 𝑢 & 𝑣 affect the oscillating behavior of the sinusoidal signals in the 𝑥 & 𝑦 directions, respectively 

12

Page 7: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

13

𝑠 𝑥, 𝑦 sin 2𝜋 𝑢 𝑥 𝑣 𝑦 , 0 𝑥, 𝑦 1

Periodic signals

• Periodic signal if there exist two positive constants 𝑋 & 𝑌 such that

– 𝑓 𝑥, 𝑦 𝑓 𝑥 𝑋, 𝑦 𝑓 𝑥, 𝑦 𝑌

• 𝛿 𝑥, 𝑦; ∆𝑥, ∆𝑦 is periodic with periods 𝑋 ∆𝑥 & 𝑌 ∆𝑦

• 𝑒 𝑥, 𝑦 𝑒 is periodic with periods 𝑋 1/𝑢 & 𝑌 1/𝑣

14

Page 8: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

Linear systems

• A system is linear if

– when the input consists of a weighted summation of several signals, the output will also be a weighted summation of the response of the system to each individual input signal

– 𝒮 ∑ 𝜔 𝑓 𝑥, 𝑦 ∑ 𝜔 𝒮 𝑓 𝑥, 𝑦

• Linear‐system assumption means that "the image of an ensemble of signals in (nearly) identical to the sum of separate images of each signal"

15

Impulse response

• Notation

– ℎ 𝑥, 𝑦; 𝜉, 𝜂 𝒮 𝛿 𝑥, 𝑦

• Output of system 𝒮 to input 𝛿 𝑥, 𝑦 𝛿 𝑥 𝜉, 𝑦 𝜂 (i.e., a point impulse located at 

𝜉, 𝜂 )

• Point‐spread function (PSF) or impulse‐response function (IRF) of system 𝒮

• The relationship between the input 𝑓 𝑥, 𝑦 and the output 𝑔 𝑥, 𝑦 for a linear system 𝒮:

16

"The PSF uniquely characterizes a linear system but should be known for every  𝑥, 𝑦; 𝜉, 𝜂 "

𝑔 𝑥, 𝑦 𝒮 𝑓 𝑥, 𝑦

𝒮 𝑓 𝜉, 𝜂 𝛿 𝑥 𝜉, 𝑦 𝜂 d𝜉d𝜂

𝒮 𝑓 𝜉, 𝜂 𝛿 𝑥, 𝑦 d𝜉d𝜂

𝑓 𝜉, 𝜂 𝒮 𝛿 𝑥, 𝑦 d𝜉d𝜂

superposition integral𝑓 𝜉, 𝜂 ℎ 𝑥, 𝑦; 𝜉, 𝜂 d𝜉d𝜂

Page 9: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

Shift invariance

• A system is shift‐invariant if

– an arbitrary translation of the input results in an identical translation in the output

– 𝑔 𝑥 𝑥 , 𝑦 𝑦 𝒮 𝑓 𝑥, 𝑦 if 𝑓 𝑥, 𝑦 𝑓 𝑥 𝑥 , 𝑦 𝑦

• If a linear‐system is shift‐invariant,

– 𝒮 𝛿 𝑥, 𝑦 ℎ 𝑥 𝜉, 𝑦 𝜂 instead of ℎ 𝑥, 𝑦; 𝜉, 𝜂

• The PSF becomes a 2D signal independent of 𝜉, 𝜂

• If a 𝒮 is a linear shift‐invariant (LSI) system,

– 𝑔 𝑥, 𝑦 𝑓 𝜉, 𝜂 ℎ 𝑥 𝜉, 𝑦 𝜂 d𝜉d𝜂

– or simply, 𝑔 𝑥, 𝑦 ℎ 𝑥, 𝑦 ∗ 𝑓 𝑥, 𝑦

17

convolution integral

convolution equation

Example

• For the followings, check the linear & shift invariance.

1) 𝑔 𝑥, 𝑦 2𝑓 𝑥, 𝑦2) 𝑔 𝑥, 𝑦 𝑥𝑦𝑓 𝑥, 𝑦

18

Page 10: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

Connections of LSI systems

• Cascade or serial connections

19

𝑔 𝑥, 𝑦 ℎ 𝑥, 𝑦 ∗ ℎ 𝑥, 𝑦 ∗ 𝑓 𝑥, 𝑦

ℎ 𝑥, 𝑦 ∗ ℎ 𝑥, 𝑦 ∗ 𝑓 𝑥, 𝑦

ℎ 𝑥, 𝑦 ∗ ℎ 𝑥, 𝑦 ∗ 𝑓 𝑥, 𝑦 associativity

cummutativityℎ 𝑥, 𝑦 ∗ ℎ 𝑥, 𝑦 ℎ 𝑥, 𝑦 ∗ ℎ 𝑥, 𝑦

𝑔 𝑥, 𝑦 𝑓 𝜉, 𝜂 ℎ 𝑥 𝜉, 𝑦 𝜂 d𝜉d𝜂 ℎ 𝜉, 𝜂 𝑓 𝑥 𝜉, 𝑦 𝜂 d𝜉d𝜂

• Parallel connections

𝑔 𝑥, 𝑦 ℎ 𝑥, 𝑦 ∗ 𝑓 𝑥, 𝑦 ℎ 𝑥, 𝑦 ∗ 𝑓 𝑥, 𝑦

ℎ 𝑥, 𝑦 ℎ 𝑥, 𝑦 ∗ 𝑓 𝑥, 𝑦 distributivity

Example

• Determine the PSF of two cascaded linear systems, each of which is with Gaussian PSF of 

ℎ 𝑥, 𝑦 𝑒 / and ℎ 𝑥, 𝑦 𝑒 / , respectively, where 𝜎

and 𝜎 are positive.

20

Page 11: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

Separable systems

• A 2D LSI system with PSF ℎ 𝑥, 𝑦 is a separable system if there exist two 1D systems with PSFs ℎ 𝑥 & ℎ 𝑦 , such that ℎ 𝑥, 𝑦 ℎ 𝑥 ℎ 𝑦

21

𝑔 𝑥, 𝑦 ℎ 𝑥, 𝑦 ∗ 𝑓 𝑥, 𝑦 𝑓 𝜉, 𝜂 ℎ 𝑥 𝜉, 𝑦 𝜂 d𝜉d𝜂

𝑓 𝜉, 𝜂 ℎ 𝑥 𝜉 ℎ 𝑦 𝜂 d𝜉d𝜂

𝑓 𝜉, 𝜂 ℎ 𝑥 𝜉 d𝜉 ℎ 𝑦 𝜂 d𝜂

𝑤 𝑥, 𝜂 ℎ 𝑦 𝜂 d𝜂

compute 𝑤 𝑥, 𝜂 for every 𝑦

compute 𝑔 𝑥, 𝑦 for every 𝑥

22

ℎ 𝑥, 𝑦 𝑒 / 𝑒 / 𝑒 / with 𝜎 = 4

Page 12: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

Stability

• A medical imaging system is (informally) stable if small inputs lead to outputs that do not diverge

• Bounded‐input bounded‐output (BIBO) stable system if

– 𝑔 𝑥, 𝑦 ℎ 𝑥, 𝑦 ∗ 𝑓 𝑥, 𝑦 𝐵 ∞ when  𝑓 𝑥, 𝑦 𝐵 ∞ for every  𝑥, 𝑦

• An LSI system is a BIBO stable system iff its PSF is absolutely integrable:

– ℎ 𝑥, 𝑦 d𝑥d𝑦 ∞

• When designing a medical imaging systems, model it using stable PSFs!

23

Fourier transform

• Recall the convolution equation:

– 𝑔 𝑥, 𝑦 𝑓 𝜉, 𝜂 ℎ 𝑥 𝜉, 𝑦 𝜂 d𝜉d𝜂

– Obtained by decomposing a signal into point impulses

• Note: 𝑓 𝜉, 𝜂 𝑓 𝑥, 𝑦 𝛿 𝑥 𝜉, 𝑦 𝜂 d𝑥d𝑦

• Now (as an alternative), decompose signal in terms of complex exponential signals:

– 𝑓 𝑥, 𝑦 𝐹 𝑢, 𝑣 𝑒 d𝑢d𝑣

• 𝑓 𝑥, 𝑦 ℱ 𝐹 𝑢, 𝑣 inverse Fourier transform

• Synthesis: sinusoidal composition of signal with strength 𝐹 𝑢, 𝑣 at different frequencies 

– 𝐹 𝑢, 𝑣 𝑓 𝑥, 𝑦 𝑒 d𝑥d𝑦

• 𝐹 𝑢, 𝑣 ℱ 𝑓 𝑥, 𝑦 Fourier transform

• Decomposition: determination of sinusoidal signal strength at each frequency

– FT allows us to separately consider the action of an LSI system on each sinusoidal frequency

24

Page 13: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

– Spectrum (of 𝑓 𝑥, 𝑦 )  𝐹 𝑢, 𝑣 𝐹 𝑢, 𝑣 𝑒 ∠ ,

• Magnitude spectrum  𝐹 𝑢, 𝑣 𝐹 𝑢, 𝑣 𝐹 𝑢, 𝑣– Power spectrum 𝐹 𝑢, 𝑣

• Phase spectrum ∠𝐹 𝑢, 𝑣 tan,

,

• Important relationship to keep in mind

– 𝑒 d𝑥d𝑦 𝛿 𝑢, 𝑣

25

Example

• ℱ 𝛿 𝑥, 𝑦 ?

• ℱ 𝑒 ?

• ℱ rect 𝑥 ?

26

Page 14: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

Basic FT pairs

𝑓 𝑥, 𝑦 𝐹 𝑢, 𝑣

1𝛿 𝑥, 𝑦𝛿 𝑥 𝑥 , 𝑦 𝑦𝛿 𝑥, 𝑦; ∆𝑥, ∆𝑦𝑒sin 2𝜋 𝑢 𝑥 𝑣 𝑦cos 2𝜋 𝑢 𝑥 𝑣 𝑦rect 𝑥, 𝑦sinc 𝑥, 𝑦comb 𝑥, 𝑦𝑒

𝛿 𝑢, 𝑣1𝑒comb 𝑢∆𝑥, 𝑣∆𝑦𝛿 𝑢 𝑢 , 𝑣 𝑣𝛿 𝑢 𝑢 , 𝑣 𝑣 𝛿 𝑢 𝑢 , 𝑣 𝑣 /2𝑗𝛿 𝑢 𝑢 , 𝑣 𝑣 𝛿 𝑢 𝑢 , 𝑣 𝑣 /2

sinc 𝑢, 𝑣rect 𝑢, 𝑣comb 𝑢, 𝑣𝑒

27

Intuitive grasp

• Point impulse

– Extremely nonuniform profile across space

– Hence, uniform frequency content in the Fourier domain (i.e., constant magnitude spectrum)

• Constant signal

– No variation in space

– Hence, unique spectrum concentrated at a specific frequency

• Slow signal variation

– Spectral content primarily concentrated at low frequencies

• Fast signal variation (e.g., at the edges of structures within an image)

– Spectral content primarily concentrated at high frequencies

28

Page 15: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

29

Displayed in log 1 𝐹 𝑢, 𝑣

FT properties

Property Signal FT

Linearity 𝑎 𝑓 𝑥, 𝑦 𝑎 𝑔 𝑥, 𝑦 𝑎 𝐹 𝑢, 𝑣 𝑎 𝐺 𝑢, 𝑣

Translation 𝑓 𝑥 𝑥 , 𝑦 𝑦 𝐹 𝑢, 𝑣 𝑒

Conjugation 𝑓∗ 𝑥, 𝑦 𝐹∗ 𝑢, 𝑣

Conjugate symmetry Real-valued 𝑓 𝑥, 𝑦 𝐹 𝑢, 𝑣 𝐹∗ 𝑢, 𝑣

Signal reversing 𝑓 𝑥, 𝑦 𝐹 𝑢, 𝑣

Scaling 𝑓 𝑎𝑥, 𝑏𝑦 𝐹 ,

Rotation 𝑓 𝑥 cos 𝜃 𝑦 sin 𝜃 , 𝑥 sin 𝜃 𝑦 cos 𝜃) 𝐹 𝑢 cos 𝜃 𝑣 sin 𝜃 , 𝑢 sin 𝜃 𝑣 cos 𝜃)

Circular symmetry Circularly symmetric 𝑓 𝑥, 𝑦 Circularly symmetric 𝐹 𝑢, 𝑣

Convolution 𝑓 𝑥, 𝑦 ∗ 𝑔 𝑥, 𝑦 𝐹 𝑢, 𝑣 𝐺 𝑢, 𝑣

Product 𝑓 𝑥, 𝑦 𝑔 𝑥, 𝑦 𝐹 𝑢, 𝑣 ∗ 𝐺 𝑢, 𝑣

Separable product 𝑓 𝑥 𝑔 𝑦 𝐹 𝑢 ∗ 𝐺 𝑢

Parseval's theorem𝑓 𝑥, 𝑦 d𝑥d𝑦 𝐹 𝑢, 𝑣 d𝑢d𝑣

30

Page 16: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

• Translation

– ℱ 𝑓 𝑥 𝑥 , 𝑦 𝑦 𝐹 𝑢, 𝑣 𝑒• ℱ 𝑓 𝑥 𝑥 , 𝑦 𝑦 𝐹 𝑢, 𝑣• ∠ℱ 𝑓 𝑥 𝑥 , 𝑦 𝑦 ∠𝐹 𝑢, 𝑣 2𝜋 𝑢𝑥 𝑣𝑦

– Constant phase change at each frequency  𝑢, 𝑣

• Conjugate symmetry

– The real part of the spectrum and the magnitude spectrum are symmetric functions, whereas the imaginary part and the phase spectrum are antisymmetric functions

• 𝐹 𝑢, 𝑣 𝐹 𝑢, 𝑣 & 𝐹 𝑢, 𝑣 𝐹 𝑢, 𝑣• 𝐹 𝑢, 𝑣 𝐹 𝑢, 𝑣 & ∠𝐹 𝑢, 𝑣 ∠𝐹 𝑢, 𝑣

• Parseval's theorem

– The total energy of a signal in the spatial domain equals its total energy in the frequency domain

– The FT (as well as its inverse) are energy‐preserving ("unit gain") transformations

31

Separability

32

𝐹 𝑢, 𝑣 𝑓 𝑥, 𝑦 𝑒 d𝑥d𝑦

𝑓 𝑥, 𝑦 𝑒 d𝑥 𝑒 d𝑦

𝑟 𝑢, 𝑦 𝑒 d𝑦

compute for every 𝑦

compute for every 𝑢

Page 17: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

33

Example

• Detectors of many medical imaging systems can be modeled as rect functions of different 

sizes & locations. Determine ℱ rect∆

,∆

.

34

Page 18: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

Example

• For two signals 𝑓 𝑥, 𝑦 sinc 𝑈𝑥, 𝑉𝑦 & 𝑔 𝑥, 𝑦 sinc 𝑉𝑥, 𝑈𝑦 , where 0 𝑉 𝑈. What is 𝑓 𝑥, 𝑦 ∗ 𝑔 𝑥, 𝑦 ?

35

Transfer function

• To determine the PSF of an LSI system, we have observed the system output to a point impulse

• Let us replace the point impulse with the complex exponential signal:

36

𝑔 𝑥, 𝑦 𝑒 ℎ 𝑥 𝜉, 𝑦 𝜂 d𝜉d𝜂

ℎ 𝜉, 𝜂 𝑒 d𝜉d𝜂

ℎ 𝜉, 𝜂 𝑒 d𝜉d𝜂 𝑒

𝐻 𝑢, 𝑣 𝑒

Output 𝐻 𝑢, 𝑣 Input

Page 19: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

– 𝐻 𝑢, 𝑣 ℎ 𝑥, 𝑦 𝑒 d𝑥d𝑦

• FT of the PSF ℎ 𝑥, 𝑦• Aka the transfer function of the LSI system 𝒮• AKa the frequency response or the optical transfer function (OTF) of system 𝒮• Uniquely characterizing the LSI system because the PSF is uniquely determined using its 

inverse FT:

– ℎ 𝑥, 𝑦 𝐻 𝑢, 𝑣 𝑒 d𝑢d𝑣

– From the convolution property: 𝐺 𝑢, 𝑣 𝐻 𝑢, 𝑣 𝐹 𝑢, 𝑣

37

Ex: Low‐pass filtering an image

– 𝑐 = the cutoff frequency

– Signal smoothing with fine signal details (that are mainly responsible for high‐frequency spectral content) being eliminated

38

𝐻 𝑢, 𝑣1, for 𝑢 𝑣 𝑐 ⟹ 𝐺 𝑢, 𝑣 𝐹 𝑢, 𝑣

0, for 𝑢 𝑣 𝑐 ⟹ 𝐺 𝑢, 𝑣 0

Page 20: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

Circular symmetry

• Circularly symmetric if

– 𝑓 𝑥, 𝑦 𝑓 𝑥, 𝑦 for every 𝜃

• Then,

– 𝑓 𝑥, 𝑦 is even in both 𝑥 & 𝑦, and 𝐹 𝑢, 𝑣 is even in both 𝑢 & 𝑣– 𝐹 𝑢, 𝑣   𝐹 𝑢, 𝑣 and ∠𝐹 𝑢, 𝑣 0

• Circularly symmetric 𝑓 𝑥, 𝑦

– 𝑓 𝑥, 𝑦 𝑓 𝑟 where 𝑟 𝑥 𝑦

– 𝐹 𝑢, 𝑣 𝐹 𝜚 where 𝜚 𝑢 𝑣

39

Hankel transform

• Hankel transform

– 𝐹 𝜚 2𝜋 𝑓 𝑟 𝐽 2𝜋𝜚𝑟 𝑟d𝑟

• 𝐽 𝑟 = the zeroth‐order Bessel function of the first kind

• 𝐽 𝑟 cos 𝑟 sin 𝜙 d𝜙

– Note: 𝐽 𝑟 cos 𝑛𝑟 𝑟 sin 𝜙 d𝜙 , 𝑛 0,1,2, …

– Notation: 𝐹 𝜚 ℋ 𝑓 𝑟

• Inverse HT

– 𝑓 𝑟 ℋ 𝐹 𝜚

– 𝑓 𝑟 2𝜋 𝐹 𝜚 𝐽 2𝜋𝜚𝑟 𝜚d𝜚

40

the nth‐order Bessel function of the first kind

Page 21: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

HT pairs

41

𝑓 𝑟 𝐹 𝜚

𝑒1𝛿 𝑟 𝑎rect 𝑟sinc 𝑟1 𝑟⁄

Scaling𝑓 𝑎𝑟

𝑒𝛿 𝜚 𝜋𝜚⁄ 𝛿 𝑢, 𝑣2𝜋𝑎𝐽 2𝜋𝑎𝜚𝐽 𝜋𝜚 2𝜚⁄2rect 𝜚 𝜋 1 4𝜚⁄1 𝜚⁄

1 𝑎⁄ 𝐹 𝜚 𝑎⁄

Jinc function

• ℋ rect 𝑟 ≡ jinc 𝜚

– rect 𝑟 = a circularly symmetric unit disk

– 𝐽 𝑟 = the first‐order Bessel function of the first kind

– The rect and jinc functions form a HT pair as the rect and sinc functions form the FT pair

42

Page 22: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

Sampling

• Discretization or sampling

– Transformation from continuous signals into collections of numbers

– Retaining representative signal values & discarding the rest

• Rectangular sampling

– 𝑓 𝑚, 𝑛 𝑓 𝑚∆𝑥, 𝑛∆𝑦 for 𝑚, 𝑛 0,1, …• ∆𝑥 & ∆𝑦 = the sampling periods in the 𝑥 & 𝑦 directions, respectively

• ∆𝑥 & ∆𝑦 = the corresponding sampling frequencies

43

Coarse sampling Fine sampling

• Given a 𝑓 𝑥, 𝑦 , what are the maximum possible values for ∆𝑥 and ∆𝑦 such that 𝑓 𝑥, 𝑦can be reconstructed from the 𝑓 𝑚, 𝑛 ? 

– Sampling with too few samples results in signal corruption called aliasing

• Higher frequencies "take the alias of" lower frequencies

44

Page 23: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

45

AliasesAliasing

I. A. Cunningham | Ch. 2. Applied linear-system theoryHandbook of Medical Imaging | SPIE | 2000

46

J. T. Dobbins III | Med. Phys. | 1995

Page 24: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

Sampling signal model

47

• Sampled signal = signal  sampling function

𝑓 𝑥, 𝑦 𝑓 𝑥, 𝑦 𝛿 𝑥, 𝑦; ∆𝑥, ∆𝑦

𝑓 𝑥, 𝑦 𝛿 𝑥 𝑚∆𝑥, 𝑦 𝑛∆𝑦

𝑓 𝑚∆𝑥, 𝑛∆𝑦 𝛿 𝑥 𝑚∆𝑥, 𝑦 𝑛∆𝑦

𝑓 𝑚, 𝑛 𝛿 𝑥 𝑚∆𝑥, 𝑦 𝑛∆𝑦

Continuous signal

Discrete signal

• Reconstruction: 𝑓 𝑚, 𝑛 ⇒ 𝑓 𝑥, 𝑦 ⇒ 𝑓 𝑥, 𝑦

48

• Let us find the relationship b/w 𝑓 𝑥, 𝑦 & 𝑓 𝑥, 𝑦 :

• The spectrum 𝐹 𝑢, 𝑣 of 𝑓 𝑥, 𝑦 is obtained by shifting the spectrum 𝐹 𝑥, 𝑦 of 𝑓 𝑥, 𝑦 to 

locations ∆

,∆

for all 𝑚 & 𝑛, adding all shifted spectra & dividing the results by ∆𝑥∆𝑦

𝐹 𝑢, 𝑣 𝐹 𝑢, 𝑣 ∗ comb 𝑢∆𝑥, 𝑣∆𝑦

𝐹 𝑢, 𝑣 ∗ 𝛿 𝑢∆𝑥 𝑚, 𝑣∆𝑦 𝑛

1∆𝑥∆𝑦

𝐹 𝑢, 𝑣 ∗ 𝛿 𝑢𝑚∆𝑥

, 𝑣𝑛

∆𝑦

1∆𝑥∆𝑦

𝐹 𝑢, 𝑣 ∗ 𝛿 𝑢𝑚∆𝑥

, 𝑣𝑛

∆𝑦

1∆𝑥∆𝑦

𝐹 𝜉, 𝜂 𝛿 𝑢𝑚∆𝑥

𝜉, 𝑣𝑛

∆𝑦𝜂 d𝜉d𝜂

1∆𝑥∆𝑦

𝐹 𝑢𝑚∆𝑥

, 𝑣𝑛

∆𝑦

Page 25: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

49

Nyquist sampling theorem

• A 2D continuous band‐limited signal 𝑓 𝑥, 𝑦 , with cutoff frequencies 𝑈 and 𝑉, can be uniquely determined from its samples 𝑓 𝑚, 𝑛 𝑓 𝑚∆𝑥, 𝑛∆𝑦 , iff the sampling periods 

∆𝑥 and ∆𝑦 satisfy ∆𝑥 and ∆𝑦

– Nyquist sampling periods:  ∆𝑥 and  ∆𝑦

– Note that aliasing‐free sampling first requires band‐limited continuous signal

50

Page 26: Signals and Systemsbml.pusan.ac.kr/LectureFrame/.../SignalSystem.std.pdf · • Signals – Mathematical functions capable of modelinga variety of physical processes – Continuous

Anti‐aliasing filters

• To avoid the aliasing artifacts, filter the continuous signal using an LSF (called the anti‐aliasing filter) and then sample w/ fewer samples

– Image could be degraded by blurring rather than aliasing artifacts

• Inherent anti‐aliasing filter

– Most medical imaging systems are integrators rather than point samplers; most imaging instruments do not sample a signal at an array of points, but integrate the continuous signal locally

– 𝑓 𝑚, 𝑛 𝑓 𝑚∆𝑥, 𝑛∆𝑦– 𝑓 𝑥, 𝑦 ℎ 𝑥, 𝑦 ∗ 𝑓 𝑥, 𝑦 𝛿 𝑥, 𝑦; ∆𝑥, ∆𝑦

51

Local integration w/ PSF  anti‐aliasing filtering

Example

• Consider a medical imaging system w/ sampling periods ∆ in both the 𝑥‐ & 𝑦‐directions. What is the highest frequency allowed in the images so that the sampling is free of aliasing? If an anti‐aliasing filter, whose PSF is modeled as a rect function, is used and we ignore all the side lobes of its transfer function, what are the widths of the rect function?

52