0% found this document useful (0 votes)
244 views35 pages

Image Deblurring

The document discusses image deblurring techniques. It introduces various causes of blurring in images and methods to address blurring, including inverse filtering, Wiener filtering, and iterative deblurring using the Landweber algorithm. Inverse filtering directly inverts the blurring filter but amplifies noise. Wiener filtering balances image recovery with noise suppression. The Landweber algorithm iteratively estimates the deblurred image to avoid issues with noise amplification seen in inverse filtering.

Uploaded by

Marco Dufresne
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
244 views35 pages

Image Deblurring

The document discusses image deblurring techniques. It introduces various causes of blurring in images and methods to address blurring, including inverse filtering, Wiener filtering, and iterative deblurring using the Landweber algorithm. Inverse filtering directly inverts the blurring filter but amplifies noise. Wiener filtering balances image recovery with noise suppression. The Landweber algorithm iteratively estimates the deblurred image to avoid issues with noise amplification seen in inverse filtering.

Uploaded by

Marco Dufresne
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 35

Image Deblurring

Introduction
Inverse filtering

Wiener filtering

Suffer from noise amplification


Tradeoff between image recovery and
noise suppression

Iterative deblurring*

Landweber algorithm
EE465: Introduction t

Introduction

Where does blur come from?

Optical blur: camera is out-of-focus


Motion blur: camera or object is moving

Why do we need deblurring?

Visually annoying
Wrong target for compression
Bad for analysis
Numerous applications
EE465: Introduction t

Application (I):
Astronomical Imaging

The Story of Hubble


Space Telescope
(HST)

HST Cost at Launch


(1990): $1.5 billion
Main mirror
imperfections due to
human errors
Got repaired in 1993
EE465: Introduction t

Restoration of HST Images

EE465: Introduction t

Another Example

EE465: Introduction t

The Real (Optical) Solution

Before the repair

After the repair

EE465: Introduction t

Application (II): Law


Enforcement

Motion-blurred license plate image

EE465: Introduction t

Restoration Example

EE465: Introduction t

Application into Biometrics

out-of-focus iris image

EE465: Introduction t

Modeling Blurring Process


Linear degradation model
x(m,n)

h(m,n)

y(m,n)

w(m, n)

h(m, n)

blurring filter

w(m, n) ~ N (0, ) additive white Gaussian noise


2
w

EE465: Introduction t

10

Blurring Filter Example

FT

h(m, n) exp(

m2 n2
2

w12 w22
H ( w1 , w2 ) exp(
)
2
2

Gaussian filter can be used to approximate out-of-focus blur

EE465: Introduction t

11

Blurring Filter Example


(Cont)

FT

h(m, n)
MATLAB code: h=FSPECIAL('motion',9,30);

H ( w1 , w2 )

Motion blurring can be approximated by 1D low-pass filter along the moving d

EE465: Introduction t

12

The Curse of Noise


x(m,n)

h(m,n)

z(m,n)

y(m,n)

w(m, n) ~ N (0, w2 )
Blurring SNR

z2
BSNR 10 log10 2
w
EE465: Introduction t

13

Image Example

BSNR=10dB
BSNR=40dB
h(m,n): 1D horizontal motion blurring [1 1 1 1 1 1 1]/7
x(m,n)

EE465: Introduction t

14

Blind vs. Nonblind


Deblurring

Blind deblurring (deconvolution):


blurring kernel h(m,n) is
unknown
Nonblind deconvolution:
blurring kernel h(m,n) is known
In this course, we only cover the
nonblind case (the easier case)
EE465: Introduction t

15

Image Deblurring

Introduction
Inverse filtering

Wiener filtering

Suffer from noise amplification


Tradeoff between image recovery and
noise suppression

Iterative deblurring*

Landweber algorithm
EE465: Introduction t

16

Inverse Filter
x(m,n)

h(m,n)

y(m,n)

hI(m,n)

blurring filter

x(m,n)

inverse filter
hcombi (m,n)

To compensate the blurring, we require

hcombi (m, n) h(m, n) h I (m, n)

I
h
(m k , n l )h(k , l ) (m, n), (m, n)

k l

H I ( w1 , w2 )

1
H ( w1 , w2 )

EE465: Introduction t

17

Inverse Filtering (Cont)


x(m,n)

h(m,n)

w(m, n)

y(m,n)

hI(m,n)

x(m,n)

inverse filter

Spatial:

x (m, n) y (m, n) h I (m, n) ( x(m, n) h(m, n) w(m, n)) h I (m, n)


Frequency:

X ( w1 , w2 ) H ( w1 , w2 ) W ( w1 , w2 )
I

X ( w1 , w2 ) Y ( w1 , w2 ) H ( w1 , w2 )
H ( w1 , w2 )
X ( w1 , w2 )

W ( w1 , w2 )
H ( w1 , w2 )

amplified noise

EE465: Introduction t

18

Image Example

motion blurred image


at BSNR of 40dB

deblurred image after


inverse filtering

Q: Why does the amplified noise look so bad?


A: zeros in H(w1,w2) correspond to poles in HI (w1,w2)
EE465: Introduction t

19

Pseudo-inverse Filter
Basic idea:
To handle zeros in H(w1,w2), we treat them separately
when performing the inverse filtering

H ( w1 , w2 )

1
| H ( w1 , w2 ) |
H ( w1 , w2 )
0
| H ( w1 , w2 ) |

EE465: Introduction t

20

Image Example

motion blurred image


at BSNR of 40dB

deblurred image after


Pseudo-inverse filtering
(=0.1)

EE465: Introduction t

21

Image Deblurring

Introduction
Inverse filtering

Wiener filtering

Suffer from noise amplification


Tradeoff between image recovery and
noise suppression

Iterative deblurring*

Landweber algorithm
EE465: Introduction t

22

Norbert Wiener (18941964)

The renowned MIT professor Norbert Wiener


was famed for his absent-mindedness. While
crossing the MIT campus one day, he was
stopped by a student with a mathematical
problem. The perplexing question answered,
Norbert followed with one of his own: "In whi
direction was I walking when you stopped me
he asked, prompting an answer from the
curious student. "Ah," Wiener declared,
"then I've had my lunch
Anecdote of Norbert Wiener

EE465: Introduction t

23

Wiener Filtering

Also called Minimum Mean Square Error (MMSE) or Least-Square (LS) fil

H mmse ( w1 , w2 )

H * ( w1 , w2 )
| H ( w1 , w2 ) |2 K
constant

Example choice of K:

w2
K 2
z

noise energy
signal energy

K=0 inverse filtering

EE465: Introduction t

24

Image Example

motion blurred image


at BSNR of 40dB

deblurred image after


wiener filtering
(K=0.01)

EE465: Introduction t

25

Image Example (Cont)

K=0.1

K=0.01

EE465: Introduction t

K=0.001

26

Constrained Least Square Filtering

Similar to Wiener but a different way of balancing the tradeoff between

H * ( w1 , w2 )
H mmse ( w1 , w2 )
| H ( w1 , w2 ) |2 | C ( w1 , w2 ) |2

0 1 0
C (m, n) 1 4 1
0 1 0

Example choice of C:

=0 inverse filtering

EE465: Introduction t

Laplacian operator

27

Image Example

=0.1

=0.01

EE465: Introduction t

=0.001

28

Image Deblurring

Introduction
Inverse filtering

Wiener filtering

Suffer from noise amplification


Tradeoff between image recovery and
noise suppression

Iterative deblurring*

Landweber algorithm
EE465: Introduction t

29

Method of Successive
Substitution

A powerful technique for finding the


roots of any function f(x)
Basic idea

Rewrite f(x)=0 into an equivalent equation


x=g(x) (x is called fixed point of g(x))
Successive substitution: xi+1=g(xi)
Under certain condition, the iteration will
converge to the desired solution
EE465: Introduction t

30

Numerical Example
f ( x) x 2 3x 2
Two roots:

x1 1, x2 2

2
x
2
2
f ( x) x 3 x 2 0 x
g ( x)
3

xi 2
successive substitution:
xi 1
3
EE465: Introduction t

31

Numerical Example (Cont)

Note that iteration quickly converges to x=1

EE465: Introduction t

32

Landweber Iteration
Linear blurring

Y ( w1 , w2 ) H ( w1 , w2 ) X ( w1 , w2 )

We want to find the root of f

( X ) Y HX

f ( X ) 0 X X f ( X ) X (Y HX ) g ( X )

relaxation parameter controls convergence property

Successive substitution:
X0 0
X n 1 X n (Y HX n )
EE465: Introduction t

33

Asymptotic Analysis
X n 1 X n (Y HX n )
X n 1 Y RX n , R I H
n

X n R iY ( I R) 1 ( I R n 1 )Y
i 0

1 r N
r
, (r 1)

r
n 0
N 1

k 1
lim
R
Y 0
Assume convergence condition: k

we have

X lim X n ( I R ) 1Y H 1Y X
n

inverse filtering
EE465: Introduction t

34

Advantages of Landweber
Iteration

No inverse operation (e.g., division)


is involved
We can stop the iteration in the
middle way to avoid noise
amplification
It facilitates the incorporation of a
priori knowledge about the signal
(X)
intoanalysis
solution
algorithm
More
detailed
is included
in EE565: Advanced Image Process

EE465: Introduction t

35

You might also like