Biomedical Engineering Reference
In-Depth Information
120
Smooth model
Accurate model
Fast model
100
80
60
40
20
0
0
200
400
600
Iterations
800
1000
1200
1400
Figure 13.
Convergence speed versus degree or smoothness.
In this case, the model stops at the entrance of the concave area and can be used
to close gaps in the shape of unconnected regions or contours.
6.2. Comparing STOP and GO with Other Deformable Models
In this section we compare the performance and advantages of the STOP
and GO active model with the classic geodesic formulation and the region-based
approach of Paragios and Deriche [5] or Zhu [12].
The first experimental results are depicted in Figure 15. This graphic shows
convergence speed with the parameters for the geodesic and STOP and GOmodels
set at their fastest. As we can observe, the results using STOP and GO have a
faster convergence than with the classic snakes. This is due to the fact that we are
removing the constraint imposed by the curvature term along the deformation. In
our case, the curvature term only has effects in the last steps of the deformation,
thereby speeding up the process.
One of the most important and well-known problems of the geodesic snakes
is a lack of convergence in the concave regions. Figure 16 shows a real problem
where this occurs. Figure 16a shows the original image segmented by experts,
and Figures 16b,c show the geodesic model segmentation and the STOP and GO
segmentation, respectively. Observe that the geodesic model is unable to get into
