| Applications des contours actifs à la trajectométrie d'un museau de souris |
Laurent GIllet
Section Microtechnique, EPFL |
Master project
February 2006 |
Abstract |
| This project objective is to follow a mouse muzzle based on an active contour region, and to implement it in a Java plugin for Image J. To accomplish this, we will define a mouse model and position it using an approach based on the snakes. |
|
Geometric model |
| A mouse being an animal moving in a 3D environment, we chose to create a 3D model (build with ellipsoids) to finally obtain a 2D shape withholding as few degree of freedom as possible and approaching the real image. |
Figure 1 - Three dimensional model |
| The 3D model being constitute of ellipsoids, its projection is a composition of ellipses.
|
Active Contour |
| For each ellipse constituting the model, we delimit a new zone, (a circle around the ellipse), with a surface equal to the ellipse one. We define an energy of area Jrégion so that the number of low intensity pixels (the mouse pixels) is maximal in the ellipse and minimal in the circle. We therefore obtain an energy function to minimize. |
Enegy function |
| Regularization energies are added to penalize the energy function if the model is in a position that a mouse would not be able to take, and to remove redundancy. |
Java Pluggin |
| The snake manual initialization on the mouse, validate completely the chosen model and the energy calculation. Indeed it fits very well to the mouse morphology, even in the most extreme cases. |
Figure 2 - Model/image adequacy |
| Even though the model results for an image are very satisfactory; they are a bit less efficient when applied frame by frame. Indeed when the optimization algorithm is applied on a fixed image, the model sets on the optimal position in a large majority of case. |
Result |
| Finally, we obtain the following result : |
|
