Biomedical Engineering Reference
In-Depth Information
(a)
(b)
Figure 5. Two failure examples of tongue image segmentation by snakes. Notice the white
curves are the boundary-finding results.
See attached CD for color version.
results for the subsequent analysis, we introduce an interactive refinement tech-
nique based on the mechanism of the snake pit.
In essence, a snake is an interactive segmentation model due to manual ini-
tialization of curves, although some researchers have attempted to improve its
intelligence through setting initial curves automatically in some special applica-
tions, such as the case in our work. In the classical snake model proposed by Kass
et al. [43], interaction can also be represented by the mechanism of the snake pit.
The snake pit was initially developed by Kass et al. in their experimental user
interface, which has proven very useful for semiautomatic image segmentation.
The interface allows a user to select starting points and exert forces on snakes
interactively as they minimize their energy. In order to specify a particular image
feature, the user has only to pull a snake near the feature. Once close enough, the
energy minimization will push the snake the rest of the way in. These two cases
are characterized with spring and volcano icons, respectively, in their interface.
The pit mechanism can allow the user to accurately control the behavior of the
snake with little effort in specifying the feature points of interest. Figure 6 gives
a geometric interpretation of the snake pit.
In our implementation, only springs are adopted to pull a snake to the specified
true boundary. According to [43], creating a spring between two points
x 1 and
x 2
k ( x 1 x 2 ) 2 to the external energy. A pair of points for the snake
pit is shown in Figure 7.
After setting such point pairs on the snakes and the desired boundaries, we
choose the existing segmentation results as initial curves, and then deform the
snakes with extra spring forces once again. Finally, we can get refined segmenta-
tion results, some examples of which are shown in Figure 8.
simply adds
Search WWH ::




Custom Search