Title: Geometric Active Contours
1Geometric Active Contours
Computer Science Department
Technion-Israel Institute of Technology
- Ron Kimmel
- www.cs.technion.ac.il/ron
Geometric Image Processing Lab
2Edge Detection
- Edge Detection
- The process of labeling the locations in the
image where the gray levels rate of change is
high. - OUTPUT edgels locations,
- direction, strength
- Edge Integration
- The process of combining local and perhaps
sparse and non-contiguous edgel-data into
meaningful, long edge curves (or closed contours)
for segmentation - OUTPUT edges/curves consistent with the local
data -
3The Classics
- Edge detection
- Sobel, Prewitt, Other gradient estimators
- Marr Hildreth
- zero crossings of
- Haralick/Canny/Deriche et al.
- optimal directional local max of derivative
- Edge Integration
- tensor voting (Rom, Medioni, Williams, )
- dynamic programming (Shashua Ullman)
- generalized grouping processes (Lindenbaum et
al.)
4The New-Wave
- Snakes
- Geodesic Active Contours
- Model Driven Edge Detection
5Geodesic Active Contours
- Snakes Terzopoulos-Witkin-Kass 88
- Linear functional efficient implementation
- non-geometric depends on parameterization
- Open geometric scaling invariant, Fua-Leclerc 90
- Non-variational geometric flow
- Caselles et al. 93, Malladi et al. 93
- Geometric, yet does not minimize any functional
- Geodesic active contours Caselles-Kimmel-Sapiro
95 - derived from geometric functional
- non-linear inefficient implementations
- Explicit Euler schemes limit numerical step for
stability - Level set method Ohta-Jansow-Karasaki 82,
Osher-Sethian 88 - automatically handles contour topology
- Fast geodesic active contours Goldenberg-Kimmel-Ri
vlin-Rudzsky 99 - no limitation on the time step
- efficient computations in a narrow band
6Laplacian Active Contours
- Closed contours on vector fields
- Non-variational models Xu-Prince 98, Paragios et
al. 01 - A variational model Vasilevskiy-Siddiqi 01
- Laplacian active contours open/closed/robust
-
Kimmel-Bruckstein 01
Most recent variational measures for good old
operators Kimmel-Bruckstein 03
7Segmentation
8Segmentation
Caselles,Kimmel, Sapiro ICCV95
9Segmentation
10Woodland Encounter Bev Doolittle 1985
- With a good prior who needs the data
11Segmentation
Caselles,Kimmel, Sapiro ICCV95
12Prior knowledge
13Prior knowledge
14Segmentation
15Segmentation
16Segmentation
Caselles,Kimmel, Sapiro ICCV95
17Segmentation
- With a good prior who needs the data
18Wrong Prior???
19Wrong Prior???
20Wrong Prior???
21Curves in the Plane
C(0.1)
C(0.2)
C(0.7)
C(0)
C(0.4)
C(0.8)
C(0.95)
y
C(0.9)
x
22Arc-length and Curvature
C
23Calculus of Variations
- Find C for which is an
extremum -
-
- Euler-Lagrange
24Calculus of Variations
- Important Example
- Euler-Lagrange , setting
-
- Curvature flow
25Potential Functions (g)
I(x,y)
I(x)
Image
x
x
g(x)
g(x,y)
Edges
x
x
26Snakes Geodesic Active Contours
- Snake model
- Terzopoulos-Witkin-Kass 88
- Euler Lagrange as a gradient descent
-
- Geodesic active contour model
- Caselles-Kimmel-Sapiro 95
- Euler Lagrange gradient descent
27Maupertuis Principle of Least Action
p
1
- Snake Geodesic active contour
- up to some , i.e
-
- Snakes depend on parameterization.
- Different initial parameterizations
- yield solutions for different
- geometric functionals
y
0
x
Caselles Kimmel Sapiro, IJCV 97
28Geodesic Active Contours in 1D
I(x)
- Geodesic active contours are
- reparameterization invariant
x
g(x)
x
29Geodesic Active Contours in 2D
G I
s
g(x)
30Controlling -max
Smoothness
g
I
Cohen Kimmel, IJCV 97
31Fermats Principle
- In an isotropic medium, the paths taken by light
rays are extremal geodesics w.r.t. -
- i.e.,
Cohen Kimmel, IJCV 97
32Experiments - Color Segmentation
Goldenberg, Kimmel, Rivlin, Rudzsky, IEEE T-IP
2001
33 Tumor in 3D MRI
Caselles,Kimmel, Sapiro, Sbert, IEEE T-PAMI 97
34Segmentation in 4D
Malladi, Kimmel, Adalsteinsson, Caselles,
Sapiro, Sethian SIAM Biomedical workshop 96
35Tracking in Color Movies
Goldenberg, Kimmel, Rivlin, Rudzsky, IEEE T-IP
2001
36Tracking in Color Movies
Goldenberg, Kimmel, Rivlin, Rudzsky, IEEE T-IP
2001
37Edge Gradient Estimators
Xu-Prince 98, Paragios et al. 01,
Vasilevskiy-Siddiqi 01, Kimmel-Bruckstein 01
38Edge Gradient Estimators
- We want a curve with large points and
small s so - Consider the functional
- Where is a scalar function, e.g.
.
39The Classic Connection
- Suppose and we consider a closed
contour for C(s). - We have
- and by Greens Theorem we have
40The Classic Connection
- Therefore
- Hence curves that maximize are
curves that enclose all regions where
is positive! - We have that the optimal curves in this case are
- The Zero Crossings of the Laplacian
- isnt this familiar?
41The Classic Connection
- It is pedagogically nice, but the MARR-HILDRETH
edge detector is a bit too sensitive. - So we do not propose a grand return to MH but a
rethinking of the functionals used in active
contours in view of this. - INDEED, why should we ignore the gradient
directions (estimates) and have every edge
integrator controlled by the local gradient
intensity alone?
42Our Proposal
- Consider functional of the form
- These functionals yield regularized curves that
combine the good properties of LZCs where
precise border following is needed, with the good
properties of the GAC over noisy regions!
43Implementation Details
- We implement curve evolution that do gradient
descent w.r.t. the functional - Here the Euler Lagrange Equations provide the
explicit formulae. - For closed contours we compute the evolved curve
via the Osher-Sethian miracle numeric level set
formulation.
44Closed contours
GAC
LZC
LZC
GAC
Kimmel-Bruckstein IVCNZ01
45Closed contours
GAC
LZC
LZCeGAC
Kimmel-Bruckstein IVCNZ01
46Open contours
- Along the curve
- b.c. at C(0) and C(L)
Kimmel-Bruckstein IVCNZ01
47Open contours
Kimmel-Bruckstein IVCNZ01
48Geometric Measures
- Weighted arc-length
- Weighted area
- Alignment
- Robust-alignment
-
- e.g.
Variational meaning for Marr-Hildreth edge
detector Kimmel-Bruckstein IVCNZ01
49Geometric Measures
- Minimal variance
- Chan-Vese, Mumford-Shah,
- Max-Lloyd, Threshold,
50Geometric Measures
51Haralick/Canny-like Edge Detector
- Haralick suggested as edge
detector
Laplace
52Haralick/Canny Edge Detector
co-area
h
Thus, indicates optimal alignment
topological homogeneity
53Closed Contours Level Set Method
y
- implicit
representation of C - Then,
- Geodesic active contour level set formulation
- Including weighted (by g) area minimization
x
C(t)
y
C(t) level set
x
54Operator Splitting Schemes
- Additive operator splitting (AOS) Lu et al. 90,
Weickert, et al. 98 - unconditionally stable for non-linear diffusion
- Given the evolution
-
- write
- Consider the operator
-
- Explicit scheme
- , the time step, is upper bounded for
stability
55Operator Splitting Schemes
- Implicit scheme
- inverting large bandwidth matrix
- First order, semi-implicit, additive operator
splitting (AOS), or - locally one-dimensional (LOD) multiplicative
schemes are stable and efficient given by linear
tridiagonal systems of equations - that can be solved for by Thomas algorithm
LOD
AOS
56Operator Splitting Schemes
- We used the following relation (AOS)
- Locally One-Dimensional scheme (LOD)
- Decoupling the axes and the implicit formulation
leads to computational efficiency - The 1st order splitting idea is based on the
operator expansion
57Example Geodesic Active Contour
y
- The geodesic active contour model
- Where I is the image and f the implicit
representation of the curve - If f is a distance, then ,
- and the short time evolution is
- Note that and thus
- can be computed once for the whole image
x
C(t)
y
x
Goldenberg, Kimmel, Rivlin, Rudzsky, IEEE T-IP
2001
58Example Geodesic Active Contour
- f is restricted to be a distance map
- Re-initialization by Sethians
- fast marching method every iteration in
- O(n).
- Computations are performed in a narrow band
around the zero set - Multi-scale approach
- process a Gaussian pyramid of the image
y
x
C(t)
y
x
59Tracking Objects in Movies
- Movie volume as a spatial-temporal 3D hybrid
space - The AOS scheme is
- Edge function derived by the
- Beltrami framework Sochen Kimmel Malladi 98
- Contour in frame n is the initial condition for
frame n1.
y
t
x
y
t
x
60Experiments - Curvature Flow
61Experiments - Curvature Flow CPU Time
62Tracking
Goldenberg, Kimmel, Rivlin, Rudzsky, ECCV 2002
63Tracking
Goldenberg, Kimmel, Rivlin, Rudzsky, ECCV 2002
64Tracking
Goldenberg, Kimmel, Rivlin, Rudzsky, ECCV 2002
65Information extraction
Goldenberg, Kimmel, Rivlin, Rudzsky, ECCV 2002
66Thin Structures
Holzman-Gazit, Goldshier, Kimmel 2003
67Segmentation in 3D
Change in topology
Caselles,Kimmel, Sapiro, Sbert, IEEE T-PAMI 97
68Gray Matter Segmentation
- Coupled surfaces
- EL equations
Goldenberg Kimmel Rivlin Rudzsky, VLSM 2001
69Gray Matter Segmentation
Goldenberg Kimmel Rivlin Rudzsky, VLSM 2001
70Gray Matter Segmentation
Goldenberg Kimmel Rivlin Rudzsky, VLSM 2001
71(No Transcript)
72Futurism
- Recognition from
- periodic motion
Dynamism of a Dog on a Leash Giacomo Balla, 1912
Eadweard Muybridge, Animals in Motion, 1887
Goldenberg, Kimmel, Rivlin, Rudzsky, ECCV 2002
73Classification (dogs cats)
walk
run
gallop
cat...
Goldenberg, Kimmel, Rivlin, Rudzsky, ECCV 2002
74Classification (dogs cats)
Goldenberg, Kimmel, Rivlin, Rudzsky, ECCV 2002
75(No Transcript)
76Classification (people)
walk
run
run45
Goldenberg, Kimmel, Rivlin, Rudzsky, ECCV 2002
77Classification (people)
Goldenberg, Kimmel, Rivlin, Rudzsky, ECCV 2002
78(No Transcript)
79Conclusions
- Geometric-Variational method for segmentation and
tracking in finite dimensions based on prior
knowledge (more accurately, good initial
conditions). - Using the directional information for edge
integration. - Geometric-variational meaning for the
Marr-Hildreth and the Haralick (Canny) edge
detectors, leads to ways to design improved ones. - Efficient numerical implementation for active
contours. - Various medical and more general applications.
- www.cs.technion.ac.il/ron
80Gray Matter Segmentation
Goldenberg Kimmel Rivlin Rudzsky, VLSM 2001
81Edge Indicator Function for Color
- Beltrami framework Color image 2D surface
- in
space - The induced metric tensor for the image surface
- Edge indicator largest eigenvalue of the
structure tensor metric. It represents the
direction of maximal
change in -
I
Y
X
82AOS
- Proof
- The whole low order splitting idea is based on
the operator expansion