You are on page 1of 38

1 Segmentierung medizinischer Bilddaten Stefan Burkhardt

5 Region based segmentation


5 Region based segmentation
Segmentation
Component/region with defined properties, e.g.
intensity, texture etc.
Connected component
Pixel based segmentation: only takes intensity etc. into
account
Now: find methods which include information about the
connectivity of the pixels region based approaches
2 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5 Region based segmentation
What is a region?
A region is.
.a connected component
All pixels/voxels within a region are pairwise connected
by a neighbourhood, usually 4 (in 2D) or 6 (in 3D)
with a defined unique feature
We need to define a feature which characterizes the
region. All elements in a region match this feature.
Finally, a region should usually match a structure in the
image
3 Segmentierung medizinischer Bilddaten Stefan Burkhardt
Intensity / grey value
All pixels with a given intensity range belongs to the
region
Texture (intensity variance in a (2k+1)x(2k+1) window)
Image gradients
5 Region based segmentation
( )
( )
( ) ( ) ( )

= =
+ +
+
=
k
k i
k
k j
y x I j y i x I
k
y x
2
2
2
,

,
1 2
1
,
Usable feature
( ) ( ) y x I
y
x
y x I , ,
|
|
|
|
.
|

\
|

=
4 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5 Region based segmentation
Variances with different window sizes
image 33
55
99 77 1111
5 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5 Region based segmentation
Intensity gradients
image Gradient image
6 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5 Region based segmentation
Combination: gradient after variance
image 33 55
99 77 1111
7 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5 Region based segmentation
Combination: variance after gradient
image 33 55
99 77 1111
8 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5 Region based segmentation
Further possible combinations
Converting the image into a vector image
This means: each pixel has an intensity value and further
features, e.g. variance or gradient information
The image I becomes e.g.:
Calculation of the largest eigen-values leads to a feature
( )
( )
( )
|
|
.
|

\
|
=
y x
y x I
y x I
,
,
,

9 Segmentierung medizinischer Bilddaten Stefan Burkhardt


5.1 Representation of regions
Representation of regions
Definition of a region R
Representation by
boundaries
Run-length codes
Quad-trees
projections
( )
( )
)
`


=
otherwise 0
, if 1
,
R y x
y x u
10 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.1 Representation of regions
Boundaries
Extract the boundary and store only
the boundary pixels
Advantage:
Efficient representation of a region
Storage in chain code
Drawback:
The region must be reconstructed
for further use.
11 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.1 Representation of regions
Run-length codes
Any region can be viewed as a sequence of
alternating strings of 0s and 1s
Just count the number of 0s, then the
number of following 1s and so on..
5
1, 1, 1, 1, 1
1, 3, 1
0, 2, 3
5
12 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.1 Representation of regions
Quad-trees
The region is enclosed in a convenient rectangular area
This area is divided into four quadrants until it is totally in
fore- or background
13 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.1 Representation of regions
Quad-trees (2)
Encoding and saving of a quadtree
Recursive
On a leaf: store 0 or 1
On a node:
-Add a (
-store recursively code of the children
-Add a )
14 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.1 Representation of regions
Quad-trees (3)
1 1
1
1
1 1
0 0
0 0 0 0
0
5: (((1100)100)10(1100))
1: ()
2: (()10())
3: ((()100)10())
4: (((1100)100)10())
1
2
3
4
5
15 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.1 Representation of regions
Projections
A 2-dimensional region can be represented by projections
Projection: sum of run-lengths code 1s along a straight
line, oriented with an angle
Useful in shape analysis
0
2
3
2
0
1 3 1 2 0
16 Segmentierung medizinischer Bilddaten Stefan Burkhardt
approaches
Region-based segmentation
How to find and segment a region?
Approaches:
Region Growing
Region Merging
Split & Merge
Hierarchical methods / pyramid linking
17 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.3 Region growing
Region growing
Based on seed points:
Give the initial region
Set by the user
Are definitively within the region to be segmented
After k iterations: we have a segmented region
Look at all neighbours of this region
Are they homogeneous with respect to the criteria
Yes: add them
No: finish region growing
18 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.3 Region growing
Region growing (2)
image seed point neighbours Add homog.
neighbours
Find neighbours Add homog.
neighbours
Find neighbours Nothing added
19 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.3 Region growing
Region growing (3)
Comparision region growing and thresholding
Region Growing
Thresholding
Over
segmentation
20 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.3 Region growing
Region growing (4)
MR of a head
Feature: minimal/maximal grey value difference max. 20
Seed point 10 iterations 40 iterations 70 iterations Result after
117 iterations
21 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.3 Region growing
Region growing (5)
MR of a head
Feature: minimal/maximal grey value difference max. 50
Seed point 40 iterations 80 iterations 200 iterations Result after
320 iterations
22 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.3 Region growing
Region growing (5)
Problem: seed point is not inside the region
Seed point Result after 7
iterations
23 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.3 Region growing
Region growing (5)
MR of a head
Feature: minimal/maximal grey value difference max. 100
Seed point 40 iterations 80 iterations 200 iterations Result after
320 iterations
24 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.4 Region merging
Region merging
Region merging: initial partition of the image into small
homogeneous regions (e.g. 2x2 pixels)
If two neighbours regions are homogeneous then merge
these regions
Continue until no homogeneous neighboured regions
are left
No user set seed points is necessary!
25 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.5 Split
Split
The whole image is the initial partition
Test with a criteria if the partition is homogeneous
If yes: do not split the partition
Otherwise: split the image, e.g. into 4 rectangles
Continue with all partitions until no further splits are
necessary
Possible problem: one region can be split into more than
one partition
26 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.5 Split
Split (2)
image 1st split 2nd split 3d split
27 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.5 Split
Split (3)
MR of a head
Feature: minimal/maximal grey value difference max. 100
1 iteration 20 iterations 100 iterations 200 iterations result
28 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.5 Split
Split (4)
Difference: 20 Difference: 50 Difference: 100
MR of a head
Segmentation for 3 different differences
29 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.5 Split
Split (5)
Advantages:
No seed-points
Fast and easy
Drawback: over-segmentation
Usually too many regions are created
30 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.6 Split & merge
Split & merge
Extension of split
First step: do a split!
Then: merge homogeneous partitions
Check all pairs of neighboured partition if they are
homogeneous with the same criteria
If yes: merge then
Finish when all partitions are homogeneous
31 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.6 Split & merge
Split & merge
image After split
merged
1st merge 2nd merge 3d merge
4th merge 5th merge 6th merge
32 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.7 Hierarchical methods
Hierarchical methods
Gauss-pyramides: hierarchical representation of one image
with different resolutions and thus sizes
Given: an image I with size 2
k
+1 2
l
+1
Creation of a series of images with different resolutions
and sizes 2
(k-i)
+1 2
(l-i)
+1
From one step to the next:
Filtering the image
Take every second pixel for the next higher stage
33 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.7 Hierarchical methods
Hierarchical methods (2)
Quelle: Jhne, Digitale Bildverarbeitung, 2002
34 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.7 Hierarchical methods
Filtering
Simply taking every second pixel is not possibly:
See the sampling theorem!
Result: noise in the image
Need for filtering the noise
Leads to a downsampling of the image
The highest frequencies must be filtered
Usually application of
Averaging (mean value)
Gauss filters
35 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.7 Hierarchical methods
Pyramid linking
Initialisation: Calculation of a Gauss-pyramid
Each 4 pixels give one value of the next higher stage by
a simple calculation of their mean value
Segmentation:
All nodes are connected with a node on the next higher
stage which has the nearest value or is located at nearest
New calculation of the properties: new calculation of the
mean values
Iteration: repeat the segmentation step und new calculation
of the properties until no further change occur
36 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.7 Hierarchical methods
Pyramid linking (2)
Quelle: Lehmann, Bildverarbeitung fr die Medizin, 1997.
37 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.7 Hierarchical methods
Pyramid linking (3)
Advantages:
Automatical determination of the minimal number of
necessary classes
Drawback:
Missing translational, rotational and skaling invariance
38 Segmentierung medizinischer Bilddaten Stefan Burkhardt
5.8 Conclusion
Conclusion
Definition of regions
Properties of regions
Intensities, texture, gradients
Representation of regions:
Boundaries, run-length codes, quad-trees, projections
Region-based approaches
Region growing, region merging, split, split&merge
Hierarchical methods:
Pyramid linking

You might also like