Bayesian Level Sets for Image Segmentation
Level set methods [Osher and Sethian, 1988] are aimed to track the
evolution of a curve, like an object boundary.
The curve propagation according to a normal speed is determined as the zero
level set of a time-varying 2-D function, obtained by solving a
Partial Differential Equation (PDE).
When the normal speed is always positive, the curve propagation is
determined by a static PDE equation giving the arrival time for any point of the plane.
A fast technique for solving this problem is the
Fast Marching algorithm [Sethian, 1996].
A key for the boundary detection problem is the definition of the propagation speed.
Two original contributions:
- Propagation speed equal to the a posteriori class probability:
Bayesian level sets
- Multi-label Fast Marching algorithm
