The aim of this project is to use the front propagation models in order to segment anatomical objects in images. It provides a very fast and precise initialization to more complicated techniques using for example partial differential equations. We use the Eikonal equation, which is the stationary case of the Hamilton-Jacobi equations.
It is faster, but monotonicity is necessary and we cannot incorporate any curvature dependent speed in the scheme. However, we have developed stopping criterion for those techniques that enables to give in a few seconds an approximate segmentation of our anatomical objects, where level-sets model take hours.
The speed of the front is usually a simple function of the images grey levels, but more adequate speed can be derived for specific applications (see the application on tubular shape extraction).
|
|
The seed points for each front are initialized manually. Each different color represents a different front.
 |
 |
Here, the seed points for each front are initialized automatically, using the minima of the gradients, as done for the watershed tranform. Each different color represents a different front.
 |
 |
The seed points for each front are initialized manually. Each different color represents a different front.
|
|
|