Image Enhancement

1
Image Enhancement

• Process Image to be more suitable than the
  original for a specific application

• Stops short of information extraction

• No general theory or measure of enhancement

• Interactive iterative

2
Types of Image Enhancement

• Spatial

• Spectral

• Temporal

3
Image Enhancement and Restoration

• Spectral Multi-spectral

4
ENHANCEMENT

• Image Sharpening/ Crispening

• Sharpen, enhance, amplify edges

• Noise in image a severe problem

Differentiation df/dx df/dy
Gradient (df/dx, df/dy) Laplacian (d2f/dx2
d2y/ dy2) Many other edge operators (1-D,
2-D)

• Unsharp Masking

• Differencing of Images

• High-Pass Filtering

5
CRISPENING
6
UNSHARP MASKING
7
(No Transcript)
8
Original
9
Horizontal blurring
10
Horizontal vertical blurring
11
(No Transcript)
12
(No Transcript)
13
Image Filtering Implementation
14
Schematic of the basic 3 x 3 kernel
PIXEL INTENSITIES
a
b
c
f
e
d
g
h
i
15
Mask used for high-boost spatial filtering. The
value of the center weight is w9A-1. With A?1
-1
-1
-1
1
-1
w
-1
X
9
-1
-1
-1
16
A 3 x 3 region of an image and various masks used
to compute the derivative at the center point.
Note that all mask coefficients sum to 0,
indicating a response of 0 in constant areas, as
expected of a derivative operator.
a) Roberts
17
A 3 x 3 region of an image and various masks used
to compute the derivative at the center point.
Note that all mask coefficients sum to 0,
indicating a response of 0 in constant areas, as
expected of a derivative operator.
-1
-1
-1
-1
0
1
0
0
0
1
0
-1
1
1
1
-1
0
1
b) Prewitt
18
A 3 x 3 region of an image and various masks used
to compute the derivative at the center point.
Note that all mask coefficients sum to 0,
indicating a response of 0 in constant areas, as
expected of a derivative operator.
-1
-2
-1
-1
0
1
0
0
0
2
0
-2
1
2
1
-1
0
1
c) Sobel
19
General Sobel
PIXEL INTENSITIES
a
b
c
f
e
d
g
h
i
Schematic of the basic 3 x 3 kernel
20
General Sobel
The Sobel algorithm operates on the full array
and evaluates S(e) 1/8 ?? (a 2b c)-(g 2h
i) ?? ?(a 2d g) - (c 2f i) ? for each
picture element. This output is a measure of the
edge components passing through the kernel and is
independent of both the polarity of the edge and,
to a large extent, its orientation.
21
Sobel Operator threshold 1
22
Sobel Operator threshold 2
23
Sobel Operator threshold 3
24
ENHANCEMENT

• Image Smoothing (Noise cleaning)
  Summing independent copies
  Neighborhood Averaging
  Uniform Weighted (2D FIR
  filters)

• Noise detection and removal Local
  average rejection (out of range)
  Selective averaging Local
  Global Region Growing
  Median Filtering

25

• Variable Radius neighborhood -noise related to
  level

• False Contour Removal Dithering
  Noise addition

26
Local Averaging
PIXEL INTENSITIES
a
b
c
f
e
d
g
h
i
Schematic of the basic 3 x 3 kernel
27
Original
28
(No Transcript)
29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
(No Transcript)
33
ENHANCEMENT

• Pseudo-color Encoding

• Maps gray levels into arbitrary colors

• Eye can discriminate thousands of colors Eye
  can discriminate 20-30 gray levels

• Mapping can involve Hue
  Saturation Intensity

• Many schemes can be implemented directly in
  color look-up tables of modern displays

34
(No Transcript)
35
Density Slicing
C1
C4
C3
C2
C1
?
?
g
g
L
L
2
1
36
Geometric interpretation of the intensity-slicing
technique.
f(x,y) Gray-level axis
(White) L
Slicing plane
li
(Black) 0
y
x
37
Reduction in Observation Time Using Color
Conventional Display
80
70
60
Percentage of Maximum Observation Time
50
40
Color Display
30
20
10
5
0
-1
-2
-3
-4
-5
-6
-7
-8
-9
-10
S/N in db
38
ENHANCEMENT

• Spatial Domain

Gray Level Mapping
Histogram Modification

• Histogram is the first order probability density
  function of pixel gray level.

?
1
39
1
?
1

• T maps gray level r into transformed gray level s

• T maps p(r) into p(s)

• Useful class of transformations are

MonotonicSingle-valued
40
ENHANCEMENT

• Histogram Equalization

• Transforms arbitrary p(r) into uniform p(s)

• Equivalent to picking K new amplitude
  quantization levels at the centers of the
  K-tiles of p(r)

• Increase the apparent contrast of the image
  - points in dense intervals of p(r) stretched
  - points in sparse intervals of p(r) compressed

41
RESTORATION
Image Restoration
42
Image Restoration

• Restoration or recovery of a image recorded
  in the presence of degradations

• Requires model of degradations

a priori model - measure degradation
a posteriori model - use specific degraded
image to model degradation

• Noise estimates are generally required

• Good restoration requires good models

43
Restoration

• Homomorphic Filtering

if f(x,y) a(x,y)b(x,y)
then log(f) log(a) log(b)
normally
a illuminationb reflectancef received
brightness

• can selectively emphasize reflectance component

• This and similar transformations useful for
  multiplicative effects

44
Image Deblurring
EXAMPLES OF MOTION (CAMERA) CAUSING IMAGE BLUR

• Photo of scene taken from a moving vehicle at
  constant speed

• Photo of the moon taken from spacecraft headed
  towards it

• Photo of sports event where cameraman Tracks
  motion of participant

• Rotational motion

45
Ideal Image Reconstruction
Original Image
F(u,v)
H(u,v)
46
Linear Motion Case

• Blurring Function

• Ideal Deblurring Function

• Deblurring Approximations

A. Use Taylor Series Expansion
B. Use u
47
ISSUES

• Realizability of H(u,v)

• Effect of Noise

48
Restoration

• Inverse Filtering

assume g(x,y) ?? f(x,y) h(x-a,y-b) da db
or G(u,v) F(u,v) H(u,v)
it would appear that F(u,v)
G(u,v)/H(u,v) f(x,y) F(u,v)

• Spatial (non-point) degradation treated
  by this method

• Problem What happens if noise present?

49
Restoration
Now G(u,v) F(u,v) N(u,v) ? H(u,v)

• Problems

Noise is intensified at zeros of H(u,v)
Noise can overwhelm signal near zeros

• Optimum restoration requires noise model

50
WITH NOISE
G(u,v) F(u,v) N(u,v) H(u,v)