数字图像处理--灰度形态学 (英文)

Mathematical Morphology
L.J. van Vliet TNW-IST Quantitative Imaging
2
The basic operations are for gray-value images are, f(x) a) Complement = gray-scale inversion b) Translation: c) Offset = gray addition: d) Multiplication = gray scaling: f(x+v) f(x) + t a f(x) f1(x) À f2(x)
L.J. van Vliet TNW-IST Quantitative Imaging
18
f
f ⊗ g(σ)
dytB f
tetB f
Segmentation: Thresholding
L.J. van Vliet TNW-IST Quantitative Imaging
19
Divide the image into objects and background
4
Local MIN filter
[ εB f ]( x ) = min f ( x + β )
β ∈B
a
f(x)
minf(a,5)
g(x)
x
minf(a,9)
Opening & Top Hat
L.J. van Vliet TNW-IST Quantitative Imaging
5
Opening (or lower-envelope): min-filter followed by max-filter.
Subtract the background
I ( x, y )
I ( x, y ) − Iˆwhite ( x, y )
7000 2000 6000 1800 1600 1400 1200 3000 1000 800 600 400 200 0 0 50 100 150 200 250 2000 5000
x
Duality opening and closing
L.J. van Vliet TNW-IST Quantitative Imaging
13
(Re-)define erosion and dilation
[ f B ]( x ) [ f • B ]( x )
[ ( f B ) ⊕ B ]( x ) [ ( f ⊕ B ) B ]( x )
8
Erosion of a function f(x) by a set B a function g(x)
[ εB f ]( x ) = ∧ f−β = min f ( x + β )
β ∈B β ∈B
[ εg f ]( x ) = min { f ( x + β ) − g ( β )}
β ∈D{ g }
f c (x)
−f ( x )
ˆ = { a a = −b, for b ∈ B } B
Opening
L.J. van Vliet TNW-IST Quantitative Imaging
11
Erosion followed by dilation
[ γg f ]( x ) = [ δg εg f ]( x )
f(x)
f(x)
g(x)
x
g(x)
x
Closing
L.J. van Vliet TNW-IST Quantitative Imaging
12
Dilation followed by erosion.
[ φg f ]( x ) = [ εg δg f ]( x )
f(x)
f(x)
g(x)
x
g(x)
4000
3000
2000
1000
0 0 50 100 150 200 250
10
(Re)define erosion and dilation
f B f ⊕B
[ εB f ]( x ) [ δB f ]( x )
and the duality relation becomes
⎡ ( f B )c ⎤ ( x ) = ⎢⎣ ⎥⎦
with
ˆ ⎤ (x) ⎡ f c ⊕B ⎢⎣ ⎥⎦
17
Texture smoothing = texture threshold
1 φ +γ [ tet B f ]( x ) = ⎡⎣ 2 ( B B )⎤ ⎦f (x)
Not idempotent for all f(x) Self-dual
f
tetB f
Smoothing: morphology vs linear
3
Local MAX filter
[ δB f ]( x ) = max f ( x − β )
β ∈B
a
f(x)
maxf(a,5)
g(x)
x
maxf(a,9)
Erosion: Local minimum filter
L.J. van Vliet TNW-IST Quantitative Imaging
2000
1000
1000
0
0 0 50 100 150 200 250
0
50
100
150
200
250
7000
6000
5000
4000
3000
2000
1000
0 0 50 100 150 200 250
Байду номын сангаас
K
I ( x, y ) Iˆwhite ( x, y )
Application: Shading correction
1 δ +ε [ dyt B f ]( x ) = ⎡⎣ 2 ( B B )⎤ ⎦f (x)
Not idempotent Self-dual: dytB –f = – dytB f
f
dytB f
Application: Smoothing 2
L.J. van Vliet TNW-IST Quantitative Imaging
e) Intersection = minimum operator: f1(x) ¿ f2(x) f) Union = maximum operator:
a)
b)
c)
d)
e)
f)
Dilation: Local maximum filter
L.J. van Vliet TNW-IST Quantitative Imaging
7000
max distance
7000
Symmetry background peak
6000
6000
5000
5000
4000
4000
3000
3000
I ( x, y )
2000
2000
1000
1000
0
0 0 50 100 150 200 250
0
50
100
150
200
250
7000
6000
5000
minf ( maxf ( f , size ), size )
f(x)
size
x
bot_hat(f , size ) = f − upp(f , size )
Dilation
L.J. van Vliet TNW-IST Quantitative Imaging
7
Dilation of a function f(x) by a set B a function g(x)
L.J. van Vliet TNW-IST Quantitative Imaging
21
f ( x, y )
⎡ δB f ⎤ ( x ) ⎣ size ⎦
⎡ εB δB f ⎤ ( x ) ⎣ size size ⎦
size = 3 size = 7 size = 13 size = 27
size = 3 size = 7 size = 13 size = 27
14
dilation
closing
erosion
opening
Ramp + Texture
L.J. van Vliet TNW-IST Quantitative Imaging
15
Morphological filters can unravel an image into ramps and textures Textures cannot be distinguished from noise
Mathematical Morphology
Introduction to functions
Lucas J. van Vliet
http://homepage.tudelft.nl/e3q6n/
1 Quantitative Imaging Group Department Imaging Science & Technology Faculty of Applied Sciences
1000 900 800 700 600 500 400 300 200 100 0 0 50 100 150 200 250
7000
max distance
7000
Symmetry background peak
6000
6000
5000