You are on page 1of 70

Image Restoration

Digital Image Processing


Content
 Introduction
 Image degradation/restoration model
 Noise models
 Restoration by spatial filtering
 Estimation of degradation functions
 Inverse filtering
 Wiener filtering
 Geometric transformation
Introduction
 Objective of image restoration
 to recover a distorted image to the original
form based on idealized models.
 The distortion is due to
 Image degradation in sensing environment
e.g. random atmospheric turbulence
 Noisy degradation from sensor noise.
 Blurring degradation due to sensors
• e.g. camera motion or out-of-focus
 Geometric distortion
• e.g. earth photos taken by a camera in a satellite
Introduction

 Enhancement  Restoration
 Concerning the  Concerning the
extraction of image restoration of
features degradation
 Difficult to quantify  Performance can be
performance quantified
 Subjective; making an  Objective; recovering
image “look better” the original image
Image degradation /
restoration model
Noise models
 Assuming degradation only due to
additive noise (H = 1)
 Noise from sensors
 Electronic circuits
 Light level
 Sensor temperature
 Noise from environment
 Lightening
 Atmospheric disturbance
 Other strong electric/magnetic signals
Noise models
 Assuming that noise is
 independent of spatial coordinates, and
 uncorrelated with respect to the image content
Noise models
Noise models
Noise models

 Other common noise models


 Rayleigh noise
 Gamma noise
 Exponential noise
 Uniform noise
Noise Models
 Rayleigh Noise
2 ( z  a )2 / b
p( z )  ( z  a)e for z  a
b
0 for z  a

 Gamma(Erlang)
Noise
 Exponential Noise
p ( z )  ae  az for z  0
=0 for z 0
Noise models

paper

salt

-3-levels
-simple constant areas
(spans from black to white)
Additive Noise

Histograms
Additive Noise

Histograms
Periodic Noise

Noise components

Periodic noise can be reduced in


via frequency domain

Are generated due to electrical or electromechanical interference during image


acquisition
Restoration by spatial
filtering

Noise is unknown

Spatial filtering is
appropriate when only
additive noise is present
Restoration by spatial
filtering S xy is the set of coordinates in
a rectangular subimage window
of size m  n centered at point
( x, y )
Restoration by spatial
filtering
Restoration by spatial
filtering
Restoration by spatial
filtering

Q is the order of filter


Restoration by spatial
filtering

Noise level is
Mean =0
Variance = 400
Restoration by spatial
filtering
 Mean filters (noise reduced by blurring)
 Arithmetic mean filter and geometric mean filter are well
suited for random noise such as Gaussian noise
 Contraharmonic mean filter is well suited for impulse noise
• Disadvantage: must know pepper noise or salt noise in
advance
Restoration by spatial
filtering
Restoration by spatial
filtering

wrong
Restoration by spatial
filtering

-- Repeated passes of median


filter tend to blur the image.
-- Keep the number of passes
as low as possible.
Restoration by spatial
filtering

Fig. 8 next
page
Restoration by spatial
filtering Pepper noise Salt noise
-High level of noise  large filter

-Median and alpha-


trimmed filter performed
better
- Alpha-trimmed did better
than median filter
Restoration by spatial
filtering
 Filters discussed so far
 Do not consider image characteristics
 Adaptive filters to be discussed
 Behaviors based on statistical characteristics
of the subimage under a filter window
 Better performance
 More complicated
 Adaptive, local noise reduction filter
 Adaptive median filter
Restoration by spatial
filtering
Restoration by spatial
filtering
Restoration by spatial
filtering
Restoration by spatial
filtering

Adaptive filtering
Restoration by spatial
filtering
Restoration by spatial
filtering

Is Z_med
impulse?

Is Z_xy
impulse?
Restoration by spatial
filtering
Periodic Noise Reduction
(Frequency Domain Filtering)

 Band-Reject Filters
 Ideal Band-reject Filter
-D(u,v) =distance from
W
H (u , v)  1 if D(u, v)  D0  the origin of the centered
2
W W freq. rectangle
0 if D0   D(u, v)  D0 
2 2 -W =width of the band
1 if D(u, v)  D0 
W -D0=Radial center of the
2 band.
Periodic Noise Reduction
(Frequency Domain Filtering)

 Butterworth Band-Reject Filter of


order n
1
H (u , v )  2n
 D (u , v)W 
1  2 2
 D (u , v )  D0 

 Gaussian Band-Reject Filter


2
1  D 2 ( u ,v )  D02 
  
2  D ( u ,v )W 
H (u, v)  1  e
Periodic Noise Reduction
(Frequency Domain Filtering)
Periodic Noise Reduction
(Frequency Domain Filtering)

 Band-Pass Filters
 Opposite operation of a band-reject fiter

H bp  1  H br (u , v)
Periodic Noise Reduction
(Frequency Domain Filtering)
 Notch Filters
 Rejects (or passes) frequencies in predefined neighborhoods
about a center frequency

Ideal
Must appear in
symmetric pairs about
the origin.

Butterworth

Gaussian
Periodic Noise Reduction
(Frequency Domain Filtering)

 Notch Filters
 Ideal
Center H (u , v)  0 if D1 (u, v)  D0 or D2 (u , v)  D0
frequency
components 1 otherwise

2 1/ 2
D1 (u, v)  (u  M / 2  u0 )  (v  N / 2  v0 ) 
2

and
2 1/ 2
D2 (u , v)  (u  M / 2  u0 )  (v  N / 2  v0 ) 
2

Shift with respect


to the center
Notch pass filter

Horizontal lines of the


noise pattern I can be
seen
Optimum Notch Filtering

Several pairs of components are present  more than just one sinusoidal component
Optimum Notch Filtering
Estimation of degradation
functions
Estimation of degradation
functions
Estimation of degradation
functions
Estimation of degradation
functions
Estimation of degradation
functions
Estimation of degradation
functions (model bases)
 Mathematical model for uniform linear motion between
the image and the sensor during image acquisition
 Let x0(t) and y0(t) denote time varying components of
motion in the x- and y-directions
 Degradation model

g ( x, y )   f  x  x0 (t ), y  y0 (t ) dt
T
where g ( x, y ) is the blurred image
0

T  j 2  ux0 ( t )  vy0 ( t ) 
G (u , v)  F (u, v)  exp dt where
0

T  j 2  ux0 ( t )  vy0 ( t ) 
H (u , v)   e dt is degradation function
0
Estimation of degradation
functions (model bases)
 Uniform Linear motion in the x and y direction
x0(t)=at/T and y0(t)=bt/T where the image has been
displaced by a total distance a in the x-direction and b
in the y-direction

T
H (u, v)  sin   (ua  vb)  e  j ( ua  vb )

 (ua  vb)
Estimation of degradation
functions (model bases)
Inverse filtering
Inverse filtering
Inverse filtering
5/6
 k ( u  M / 2)2  ( v  N / 2)2 
Degradation function H (u , v)  e

Curtain of noise
Cutting off values of the ratio outside a radius of 40, 70,85.
Wiener filtering
Wiener filtering

H (u , v ) is the degradation function


Wiener filtering

White noinse
Wiener filtering
Geometric
transformations
 Objective: to eliminate geometric
distortion that occurs when an image is
captured
 Examples of geometric distortion
 Pincushion distortion (associated with
zoom lenses)
Geometric
transformations
Geometric
transformations
 Two steps in geometric
transformation
 Spatial transformation: rearrangement
of pixels on the image plane
 Gray-level interpolation: assignment of
gray levels to pixels in the spatially
transformed image
Geometric
transformations
Geometric
transformations
 • Solution
 To formulate the spatial relocation of pixels by the
use of the corresponding tiepoints
 Tiepoints: a subset of pixels whose locations in the
input (distorted) and output (restored) images are
known.
Geometric
transformations
Geometric
transformations
Geometric
transformations
Geometric
transformations
Geometric
transformations

You might also like