Escolar Documentos
Profissional Documentos
Cultura Documentos
Fuzzy sets
Fuzzy Techniques for Image Segmentation Fuzzy sets
2 Fuzzy sets
Fuzzy image Fuzzy image
processing processing
Fuzzy Laszl
o G. Ny
ul Fuzzy 3 Fuzzy image processing
connectedness connectedness
Fuzzy thresholding
Department of Image Processing and Computer Graphics Fuzzy clustering
University of Szeged
4 Fuzzy connectedness
2008-07-12 Theory
Algorithm
Variants
Applications
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
Dealing with imperfections Image
Segmentation
Fuzzy systems
aszl
L o G. Ny
ul aszl
L o G. Ny
ul
Outline Outline
Fuzzy systems
Aoccdrnig to a rscheearch at Cmabrigde Uinervtisy, it deosnt Fuzzy systems
Fuzzy sets
mttaer in waht oredr the ltteers in a wrod are, the olny Fuzzy sets
Fuzzy image iprmoetnt tihng is taht the frist and lsat ltteer be at the rghit Fuzzy image Fuzzy systems and models are capable of representing
processing processing
pclae. The rset can be a toatl mses and you can sitll raed it diverse, inexact, and inaccurate information
Fuzzy Fuzzy
connectedness wouthit porbelm. Tihs is bcuseae the huamn mnid deos not connectedness
Fuzzy logic provides a method to formalize reasoning when
raed ervey lteter by istlef, but the wrod as a wlohe.
According to a researcher (sic) at Cambridge University, it dealing with vague terms. Not every decision is either true
doesnt matter in what order the letters in a word are, the only or false. Fuzzy logic allows for membership functions, or
important thing is that the first and last letter be at the right degrees of truthfulness and falsehoods.
place. The rest can be a total mess and you can still read it
without problem. This is because the human mind does not
read every letter by itself but the word as a whole.
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
Membership function examples Image
Segmentation
Application area for fuzzy systems
aszl
L o G. Ny
ul aszl
L o G. Ny
ul
Outline Outline
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
Object characteristics in images Image
Segmentation
Fuzzy set
aszl
L o G. Ny
ul aszl
L o G. Ny
ul
Let X be the universal set.
Outline Outline
Fuzzy sets
is concerned with occurence of events Fuzzy sets
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
Fuzzy membership functions Image
Segmentation
Fuzzy set properties
aszl
L o G. Ny
ul aszl
L o G. Ny
ul Height
height(A) = sup {A (x) | x X }
Outline Outline
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
Properties of fuzzy relations Image
Segmentation
Fuzzy image processing
aszl
L o G. Ny
ul aszl
L o G. Ny
ul
Fuzzy image processing is the collection of all approaches that
Outline is reflexive if Outline
understand, represent and process the images, their segments
Fuzzy systems x X (x, x) = 1 Fuzzy systems
and features as fuzzy sets. The representation and processing
Fuzzy sets Fuzzy sets
Fuzzy image
is symmetric if Fuzzy image
depend on the selected fuzzy technique and on the problem to
processing processing be solved.
Fuzzy
Fuzzy x, y X (x, y ) = (y , x) thresholding (From: Tizhoosh, Fuzzy Image Processing, Springer, 1997)
connectedness Fuzzy clustering
Fuzzy
is transitive if connectedness ... a pictorial object is a fuzzy set which is specified by some
[ membership function defined on all picture points. From this
x, z X (x, z) = (x, y ) (y , z) point of view, each image point participates in many
y X
memberships. Some of this uncertainty is due to degradation,
is similitude if it is reflexive, symmetric, and transitive but some of it is inherent... In fuzzy set terminology, making
figure/ground distinctions is equivalent to transforming from
Note: this corresponds to the equivalence relation in hard sets. membership functions to characteristic functions.
(1970, J.M.B. Prewitt)
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
Fuzzy image processing Image
Segmentation
Fuzzy thresholding
aszl
L o G. Ny
ul aszl
L o G. Ny
ul
Outline Outline
0 if f (x) < T1
g (x) if T1 f (x) < T2
Fuzzy systems Fuzzy systems
Fuzzy sets Fuzzy sets g (x) = 1 if T2 f (x) < T3
Fuzzy image Fuzzy image
g (x) if T3 f (x) < T4
processing processing
Fuzzy Fuzzy
thresholding thresholding
0 if T4 f (x)
Fuzzy clustering Fuzzy clustering
Fuzzy Fuzzy
connectedness connectedness
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
Fuzzy thresholding Image
Segmentation
Fuzziness and threshold selection
aszl
L o G. Ny
ul Example aszl
L o G. Ny
ul
Outline Outline
Fuzzy Fuzzy
connectedness connectedness
Outline Outline
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
k-means clustering Image
Segmentation
k-means clustering
aszl
L o G. Ny
ul Algorithm aszl
L o G. Ny
ul Issues
Outline Outline
1 Consider a set of n data points (feature vectors) to be
Fuzzy systems Fuzzy systems
clustered.
Fuzzy sets Fuzzy sets
Fuzzy image
2 Assume the number of clusters, or classes, k, is known. Fuzzy image
processing
Fuzzy
2 k < n. processing
Fuzzy
How to initialize?
thresholding thresholding
Fuzzy clustering 3 Randomly select k initial cluster center locations. Fuzzy clustering What objective function to use?
Fuzzy Fuzzy
connectedness 4 All data points are assigned to a partition, defined by the connectedness What distance to use?
nearest cluster center. Robustness?
5 The cluster centers are moved to the geometric centroid What if k is not known?
(center of mass) of the data points in their respective
partitions.
6 Repeat from (4) until the objective function is smaller
than a given tolerance, or the centers do not move to a
new point.
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
Fuzzy c-means clustering Image
Segmentation
Fuzzy c-means clustering
aszl
L o G. Ny
ul aszl
L o G. Ny
ul Algorithm
1 Consider a set of n data points to be clustered, xi .
Outline A partition of the observed set is represented by a c n Outline
2 Assume the number of clusters (classes) c, is known. 2 c < n.
Fuzzy systems matrix U = [uik ], where uik corresponds to the Fuzzy systems
3 Choose an appropriate level of cluster fuzziness, m R>1 .
Fuzzy sets
membership value of the k th element (of n), to the i th Fuzzy sets
4 Initialize the (n c) sized membership matrix U to random
Fuzzy image Fuzzy image Pc
processing cluster (of c clusters). processing values such that uij [0, 1] and j=1 uij = 1.
Fuzzy Fuzzy
Pn m
thresholding Each element may belong to more than one cluster but its thresholding (uij ) xi
Fuzzy clustering Fuzzy clustering 5 Calculate the cluster centers cj using cj = Pi=1 n m , for
Fuzzy
overall membership equals one. Fuzzy i=1 (uij )
connectedness connectedness j = 1 . . . c.
The objective function includes a parameter m controlling
(j)
Calculate the distance measures dij =
xi cj
, for all clusters
6
the degree of fuzziness.
j = 1 . . . c and data points i = 1 . . . n.
The objective function is
7 Update the fuzzy membership matrix U according to dij . If
2 1
c X
n
2 Pc dij m1
dij > 0 then uij = . If dij = 0 then the data
X
(j)
J= (uij )m
xi cj
k=1 dik
j=1 i=1 point xj coincides with the cluster center cj , and so full
membership can be set uij = 1.
8 Repeat from (5) until the change in U is less than a given
tolerance.
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
Fuzzy c-means clustering Image
Segmentation
Basic idea of fuzzy connectedness
aszl
L o G. Ny
ul Issues aszl
L o G. Ny
ul
Outline Outline
Fuzzy image
Computationally expensive Fuzzy image local hanging-togetherness
processing processing
Fuzzy Highly dependent on the initial choice of U (affinity) based on similarity
thresholding Fuzzy
Fuzzy clustering
If data-specific experimental values are not available, connectedness
Theory
in spatial location as well as
Fuzzy
connectedness m = 2 is the usual choice Algorithm
Variants
in intensity(-derived features)
Applications
Outline
relation in Z n and assigns a value to a pair of spels (c, d) Outline
Fuzzy spel affinity is a reflexive and symmetric fuzzy relation
Fuzzy systems based on how close they are spatially. Fuzzy systems
in Z n and assigns a value to a pair of spels (c, d) based on
how close they are spatially and intensity-based-property-wise
Fuzzy sets
Example Fuzzy sets
Fuzzy Fuzzy
connectedness
1 connectedness (c, d) = h( (c, d), f (c), f (d), c, d)
Theory
if kc dk < a small distance Theory
Algorithm (c, d) = kc dk Algorithm
Variants
Applications
0 otherwise
Variants
Applications
Example
Fuzzy digital space (c, d) = (c, d) (w1 G1 (f (c) + f (d)) + w2 G2 (f (c) f (d)))
(Z n , ) !
Scene (over a fuzzy digital space) 1 (x mj )2
where Gj (x) = exp
2 j2
C = (C , f ) where C Z n and f : C [L, H]
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
Paths between spels Image
Segmentation
Strength of connectedness
aszl
L o G. Ny
ul aszl
L o G. Ny
ul
Fuzzy Fuzzy
N (pcd ) = min (cj , cj+1 )
j=1,...,m1
connectedness Let Pcd denote the set of all possible paths pcd from c to d. connectedness
Theory Theory
Algorithm Then the set of all possible paths in C is Algorithm
The fuzzy -connectedness in C (K ) is a fuzzy relation in C
Variants Variants
Applications Applications
[ and assigns a value to a pair of spels (c, d) that is the
PC = Pcd
maximum of the strengths of connectedness assigned to all
c,dC
possible paths from c to d (global hanging-togetherness).
Outline Outline
Let K be the following binary (equivalence) relation in C The fuzzy object O (o) of C containing o is
Fuzzy systems Fuzzy systems
(o) = {c C | K (o, c) } where assigns an objectness value to each spel perhaps based
on f (c) and K (o, c).
Practical computation of FC relies on the following equivalence
Fuzzy connected objects are robust to the selection of seeds.
O (o) = (o)
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
Fuzzy connectedness as Image
Segmentation
Computing fuzzy connectedness
aszl
L o G. Ny
ul a graph search problem aszl
L o G. Ny
ul Dynamic programming
Outline Outline
Algorithm
Fuzzy systems Fuzzy systems
Input: C, o C ,
Fuzzy sets Fuzzy sets Output: A K-connectivity scene Co = (Co , fo ) of C
Fuzzy image Fuzzy image Auxiliary data: a queue Q of spels
processing processing
Fuzzy
Spels graph nodes Fuzzy begin
connectedness
Spel faces graph edges connectedness set all elements of Co to 0 except o which is set to 1
Theory Theory
Algorithm Algorithm
push all spels c Co such that (o, c) > 0 to Q
Variants Fuzzy spel-affinity relation edge costs Variants while Q 6= do
Applications Applications remove a spel c from Q
Fuzzy connectedness all-pairs shortest-path problem fval maxdCo [min(fo (d), (c, d))]
if fval > fo (c) then
Fuzzy connected objects connected components fo (c) fval
push all spels e such that (c, e) > 0 fval > fo (e) fval > fo (e) and (c, e) > fo (
endif
endwhile
end
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
Computing fuzzy connectedness Image
Segmentation
Brain tissue segmentation
aszl
L o G. Ny
ul Dijkstras-like aszl
L o G. Ny
ul FSE
Outline
Algorithm Outline
Input: C, o C ,
Fuzzy systems Fuzzy systems
Output: A K-connectivity scene Co = (Co , fo ) of C
Fuzzy sets Auxiliary data: a priority queue Q of spels Fuzzy sets
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
FC with threshold Image
Segmentation
FC with threshold
aszl
L o G. Ny
ul MRI aszl
L o G. Ny
ul CT and MRA
Outline Outline
Fuzzy Fuzzy
connectedness connectedness
Theory Theory
Algorithm Algorithm
Variants Variants
Applications Applications
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
Fuzzy connectedness variants Image
Segmentation
Scale-based affinity
aszl
L o G. Ny
ul aszl
L o G. Ny
ul
Outline Outline
Fuzzy sets
Multiple seeds per object Fuzzy sets
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
Object scale Image
Segmentation
Computing object scale
aszl
L o G. Ny
ul aszl
L o G. Ny
ul
Algorithm
Outline
Object scale in C at any spel c C is the radius r (c) of the Outline Input: C, c C , W , [0, 1]
Fuzzy systems largest hyperball centered at c which lies entirely within the Fuzzy systems Output: r (c)
Fuzzy sets same object region Fuzzy sets
begin
Fuzzy image Fuzzy image
processing processing
k1
while FOk (c) do
Fuzzy Fuzzy k k +1
connectedness connectedness
Theory Theory
endwhile
Algorithm Algorithm r (c) k
Variants Variants
Applications Applications
end
Outline Outline
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
kNN vs. VSRFC Image
Segmentation
Image segmentation using FC
aszl
L o G. Ny
ul aszl
L o G. Ny
ul
MR
Outline Outline
brain tissue, tumor, MS lesion segmentation
Fuzzy systems Fuzzy systems
MRA
Fuzzy sets Fuzzy sets
vessel segmentation and artery-vein separation
Fuzzy image Fuzzy image
processing processing CT bone segmentation
Fuzzy Fuzzy
connectedness connectedness
kinematics studies
Theory Theory measuring bone density
Algorithm Algorithm
Variants Variants stress-and-strain modeling
Applications Applications
CT soft tissue segmentation
cancer, cyst, polyp detection and quantification
stenosis and aneurism detection and quantification
Digitized mammography
detecting microcalcifications
Craniofacial 3D imaging
visualization and surgical planning
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
Protocols for brain MRI Image
Segmentation
FC segmentation of brain tissues
aszl
L o G. Ny
ul aszl
L o G. Ny
ul
Outline Outline
1 Correct for RF field inhomogeneity
Fuzzy systems Fuzzy systems 2 Standardize MR image intensities
Fuzzy sets Fuzzy sets
3 Compute fuzzy affinity for GM, WM, CSF
Fuzzy image Fuzzy image
processing processing 4 Specify seeds and VOI (interaction)
Fuzzy Fuzzy
connectedness connectedness 5 Compute relative FC for GM, WM, CSF
Theory Theory
Algorithm Algorithm 6 Create brain intracranial mask
Variants Variants
Applications Applications
7 Correct brain mask (interaction)
8 Create masks for FC objects
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
Brain tissue segmentation Image
Segmentation
MS lesion quantification
aszl
L o G. Ny
ul SPGR aszl
L o G. Ny
ul FSE
Outline Outline
Fuzzy Fuzzy
connectedness connectedness
Theory Theory
Algorithm Algorithm
Variants Variants
Applications Applications
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
Brain tumor quantification Image
Segmentation
Skull object from CT
aszl
L o G. Ny
ul aszl
L o G. Ny
ul
Outline Outline
Fuzzy Fuzzy
connectedness connectedness
Theory Theory
Algorithm Algorithm
Variants Variants
Applications Applications
Fuzzy Fuzzy
Techniques for Techniques for
Image
Segmentation
MRA slice and MIP rendering Image
Segmentation
MRA vessel segmentation and
aszl
L o G. Ny
ul aszl
L o G. Ny
ul artery/vein separation
Outline Outline
Fuzzy Fuzzy
connectedness connectedness
Theory Theory
Algorithm Algorithm
Variants Variants
Applications Applications