Semi-automatic teeth segmentation in 3D models of dental casts using a hybrid methodology - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 367

FIGURE 5 Some important objects are obtained by the hybrid technique. Tree (a), tall post
(b), road sign (c).
• Two input parameters are set up prior to execution: support size and angular resolution.
• Then, we use NARF as a descriptor to obtain the target landmarks;
• The final step is to segment the cloud using each landmark as a seed in the Min-Cut all,
gorithm.
In this work, the hybrid technique is based in NARF, Min-Cut, and Region Growing [ 20 ] .
The methodology can be summarized as shown below:
• First of all, the gum is separated from teeth using the Region Growing method.
• Once gum is set apart, a set of landmarks is extracted using the NARF technique in the seg-
mented teeth region.
• Subsequently, each landmark is used as a source in the Min-Cut method.
The final segmentation is composed of the gum and a set of tooth regions. Notice that, un-
like the case of outdoor scene segmentation, there is no need to preestablish angular resolution
and support size. This is because overall teeth geometry does not vary greatly from one model
to another.
4 Results of segmentation techniques applied to 3D
dental models
Exploration results obtained by applying different segmentation techniques described in Sec-
tion 3 are shown in this section. Five 3D dental models were used in this analysis ( Figure 10 ).
Each cloud has approximately 50,000 points.
The tests were performed on all models; some important conclusions drawn from this ana-
lysis are mentioned. Finally, a semi-automatic segmentation methodology for 3D dental mod-
els is presented.
4.1 First, a Test Using RANSAC
As mentioned in Section 3.1 , RANSAC allows automatic recognition of known forms of the
data (planes, cylinders, spheres, and tori). In this case a planar model was used, with a
threshold of ± 5 mm, as shown in Figure 6(a) . The result obtained applying this technique is
shown in Figure 6(b) . The points retain colors red (dark gray in the print version) and blue
(light gray in the print version) according to the region they have been assigned to. Notice that