Title: Geodesic Active Contours
1Geodesic Active Contours
2Introduction
- Proposed by
- Vincent Caselles
- Universitat Pompeu Fabra (Spain)
- http//www.iua.upf.es/vcaselles/
- Ron Kimmel
- Technion University (Israel)
- http//www.cs.technion.ac.il/ron/
- Guillermo Sapiro
- University of Minnesota
- http//www.ece.umn.edu/users/guille/
- Geodesic Active Contours.
- International Journal of Computer Vision, Vol.
22, No. 1, - pp 61-79, 1997
?
?
3Energy based active contours
- Classical active contour models
- deforming initial contour C0 towards the boundary
of the object to be detected - First two terms control the smoothness of
contours (internal energy), Third term is
responsible for attracting contour towards edge
(external energy) - Solving this equation involves in finding C that
minimize E
4Energy based active contours
- Let ß 0,
- Regularization effect on the geodesic curve comes
from curvature based curve flows - Experiments show this curvature driven curve
motions to be very effective - To derive the relation between this model and
geometric curve evolution ones.
5Energy based active contours
- Let g(r) be steadily decreasing function,
- g(r) ? 0 as r ? 8
- Drawbacks
- The approach is not topology independent.
- Too many parameters
- When considering more than one object
6Geodesic Curve Flow
7Geodesic
- Definition
- A geodesic is a locally length-minimizing curve
by mathworld, http//mathworld.wolfram.com
8Geodesic
- Property
- The shortest distance between two points is the
length of a geodesic between the points - Examples
- In the plane, the geodesics are straight lines
- On the sphere, the geodesics are great circles
(like the equator)
9Riemannian Geometry
- Metric gij
- The length of a segment of a curve parameterized
by t, - from a to b, is defined as
- Ex) A 2D Euclidean metric gij
- Curvature k
- The amount by which an geometric object deviates
from being flat, more deviates bigger k. - Ex) For a plane curve C(t) (x(t), y(t)), the
curvature k is -
- where the dot denote
differentiation respect to t
10Geodesic Curve Flow
11Maupertuis Principle
If
Curves C(q) is in Euclidean space which are
extremal corresponding to the Hamiltonian H ,
and have a fixed energy level E0(law of
conservation of energy)
then
Curves C(q) are geodesics, with non-natural
parameter, with respect to the new metric
12Geodesic Curve Flow
- The Maupertuis Principle explains,
Minimizing E(C) with H E0,
Minimizing (C (C1, C2)), finding geodesic curve
measured by gij
Let E0 Eint-Eext 0,
13Geodesic Curve Flow
The Problem of minimizing energy
Min
The Problem of geodesic computation in a
Riemannian space, according to a new metric
14Geodesic Curve Flow
- In order to minimize
- we should follow the curve evolution equation
- (a way of minimizing via the steepest-descent
method)
This differential equation is derived by the
Euler-Lagrange of
NOTE
15Geodesic Curve Flow
- Curve evolution equation
- Shows how each point in the active contour C
should move in order to decrease the length.
16The Level-set Geodesic Flow
- To find the geodesic curve
- We compute steepest-descent flow
- But, the geodesic flow is represented using the
level-set approach
17Level Set Representation
- C is level set function u.
- Level Set
- A set points of u C
- Especially, if u0, C is zero level set
- Parameter free, Topology free
- Accurate, Stable
- ?
- Moving to normal direction
- Topological change of C(t)
- can be handled
18Level Set Representation
- Geodesic problem is equivalent to searching for
the steady state solution(?u/?t0) of the
following evolution equation (u(0,C)u0(C) ) - This equation is the Euler-Lagrange of
- with C represented by a level-set of u, and the
curvature is computed on the level-sets of u
Geodesic computation
19250years ago
20Maupetuis Principle
The stream engine
An energy minimization problem is equivalent to
finding a geodesic curve in a Riemannian space
21150years ago
22The Civil war
The solution is given by a geodesic curve in a
Riemannian space.
2333years ago
24Lena Image
The Playboy Nov 1972
The Lennas issue is Playboy's best selling
issue ever
2517years ago
26Active contour models
The 88 Seoul Olympic
Solving this equation involves in finding Curve
that minimize Energy
2710years ago
28Geodesic Active Contours
MS Windows 95
Solving this equation involves in finding Curve
that minimize Energy
29The existing model
- Caselles et al.(93) proposed the following model
for boundary detection - means that each one of
the level-sets C of u is evolving according to
Geometric smoothing
Constant velosity similar to the balloon force
30Curvature Flow
- Curvature flow
- The flow decreases the total curvature
- geometric smoothing property
- smoothing
- shortening
31Constant Velocity
- The balloon force
- An area minimizing force
- The contour will propagate inwards by
minimization of the interior - Its crucial in order to capture non-convex
shapes small gaps
32The stopping function
- Back to energy based active contours
- g(I) The stopping function
- The main goal is actually to stop the evolving
curve when it arrives to objects boundaries.
Internal energy
External energy
33Return to Geodesic framework
- is naturally incorporated in
geodesic framework - This term attracts the curve to the boundaries of
the objects. - ?g points toward the middle of the boundaries
- We can remove the constant velocity term with the
aid of ?g ?u.
34Boundary Detection
- It will lead the propagating curve into the
boundary and force it to stay there. - Its possible to detect boundaries with high
differences in their gradient values.
35Constant velocity
- We can add constant velocity term to Geodesic
framework - To increase the speed of convergence
- To help to avoid certain local minima
36(Almost) Done
- means that the level-sets move according to
- This is the level-sets representation of the
modified solution of the following geodesic
problem
37Geodesic active contour
This page is from sapiros presentation
38Geodesic Model
- The solution of the geodesic active contours
model satisfy the existence and uniqueness. - Based on the theory of viscosity solutions
39Experimental Results
- The characteristics of this image
- Separated by only a few pixels
- Shadows
- But, stronger attraction force provided by the
term - towards the real boundaries.
- Inward motion to detect two objects
40Experimental Results
- The characteristics of this image
- Separated by only a few pixels
- Shadows
- Outward motion to detect two objects
41Experimental Results
- Tumor detection
- Inward geodesic flow
- The difficulties are caused by the triangular
shaped portion at the top left part.
42Experimental Results
- The geometric model without new gradient term
- Very sensitive to variation of the gradient along
the object boundaries and the noise in the image - Curve did not stop at the correct position
- So, it needs more complicated stopping conditions
- Use A-priori knowledge
- In geodesic model on the other hand, the stopping
is obtained automatically and new gradient term
solves many hard problems.
43Experimental Results
44Experimental Results
- 3D extension of the geodesic flow
- The computation of minimal surfaces
45Concluding Remarks
- A geodesic formulation for active contours
- The connections between classical energy based
contours and geometric curve evolution - Introduce a new term to the curve evolution
models - Improve the detection of boundaries with large
differences in their gradient - Free the model from the need to estimate crucial
parameters - Active contour approach with topology independent
46(No Transcript)