Bio-Medical Materials and Engineering 24 (2014) 3303–3309 3303

DOI 10.3233/BME-141153
IOS Press

Individual tooth region segmentation using

modified watershed algorithm with
morphological characteristic
Sung Dae Na a, GihyounLee a, Jyung Hyun Lee band Myoung Nam Kim c,*
a
Department of Medical & Biological Engineering, Graduate School, Kyungpook National University,
Daegu, Republic of Korea
b
Department of Biomedical Engineering, Kyungpook National University Hospital, Daegu, Republic
of Korea
c
Department of Biomedical Engineering, School of Medicine, Kyungpook National University, Daegu,
Republic of Korea

Abstract. In this paper, a new method for individual tooth segmentation was proposed. The proposed method is composed of
enhancement and extraction of boundary and seed of watershed algorithm using trisection areas by morphological characte-
ristic of teeth. The watershed algorithm is one of the conventional methods for tooth segmentation; however, the method has
some problems. First, molar region detection ratio is reduced because of oral structure features that is low intensities in molar
region. Second, inaccurate segmentation occurs in incisor region owing to specular reflection. To solve the problems, the
trisection method using morphological characteristic was proposed, where three tooth areas are made using ratio of entire
tooth to each tooth. Moreover, the enhancement is to improve the intensity of molar using the proposed method. In addition,
boundary and seed of watershed are extracted using trisection areas applied other parameters each area. Finally, individual
tooth segmentation was performed using extracted boundary and seed. Furthermore, the proposed method was compared with
conventional methods to confirm its efficiency. As a result, the proposed method was demonstrated to have higher detection
ratio, better over segmentation, and overlap segmentation than conventional methods.

Keywords: Segmentation, tooth, watershed, oral scanner

1. Introduction

Tooth healthcare quality has increased recently because of graying improvements and a higher stan-
dard of living. Existing problems in tooth healthcare are related to the healing site for oral disease, ob-
stacles, and aesthetics. Prosthetics is an aspect of tooth healthcare that requires considerable time and
high cost throughout the process, from impression acquisition to final manufacture [1,2]. Therefore,
researchers attempt to address these problems with an image-based intra-oral-scanner [3–5]. In addi-
tion, tooth region detection requires research into dental optics-image devices, and accurate detection

*
Corresponding author: Myoung Nam Kim, Department of Biomedical Engineering, School of Medicine, Kyungpook Na-
tional University, Daegu, Republic of Korea. Tel.: +82-53-200-5266; Fax: +82-53-200-5364; E-mail: [email protected].

0959-2989/14/$27.50 © 2014 – IOS Press and the authors.

This article is published with Open Access and distributed under the terms of the Creative Commons Attribution and Non-Commercial License.
3304 S.D. Na et al. / Individual tooth region segmentation using modified watershed algorithm

(a) (b)

(c) (d)
Fig. 1. Preprocessing result, (a) G of RGB space image, (b) R of RGB space image, (c) tooth complement image and (d)
tooth emphatic image.

of the tooth image requires that there should be no over-or overlap-segmentation from the incisor to
the molar. Watershed algorithm is a suitable conventional method to segment the individual tooth re-
gions. However, a detection ratio of watershed algorithm was reduced because of the image intensity
in the molar region, both the tooth detection ratio and the accuracy decrease owing to the use of So-
bel’s edge and the regional maximum method [6]. In order to solve the problems, the trisection method
of entire tooth area using the morphological characteristic of teeth was proposed. Moreover, the boun-
dary and seed of watershed are made using trisection areas applied each parameters to overcome the
reduced detection ratio and inaccurate rate.
The proposed methods have the following steps. First, preprocessing was performed to remove un-
necessary parts of the gingival and muscles around the tooth, except for the tooth on the color tooth
image. Second, three tooth areas are made using morphological characteristic of teeth to extract the
regions of interest. Third, image enhancement was performed to improve the intensity in the molar
region using the proposed method. Next, watershed seed and boundary are extracted using trisection
areas applied other parameters on each areas. Finally, the watershed algorithm using the acquired seed
and boundary perform the individual tooth segmentation.

2. Method

2.1. Preprocessing

Emphatic tooth image was obtained in preprocessing using the RGB (red, green, blue) space from
the difference on the gingival and oral tooth muscle. The tooth regions similarly appear on a histogram
of the R, G, and B on the RGB space; however, a histogram of the R intensity appears, except for
tooth regions. In summary, the complement and emphatic images of the tooth were acquired using the
RGB space difference. Figure 1 shows the results of the experiment using the difference in the RGB
space [6].
 S.D. Na et al. / Individual tooth region segmentation using modified watershed algorithm 3305

2.2. Proposed trisection method using morphological characteristic and enhancement

The intensities of molar regions are low on the tooth image because of the tooth structure. Different
intensities cause reductions in tooth detection. In order to improve the intensities, the trisection method
was utilized to divide tooth area according to teeth width.
The proposed trisection method has the following steps. First, binary image from the acquired em-
phatic tooth image in the preprocessing was made with the applied threshold of 25. Then, unnecessary
information was removed from the rest of the image using component size. Second, four pixels are
extracted using the acquired binary image, and entire tooth region is made to connect the pixels. Eq.
(1) presents the relation of the entire tooth area and the result of preprocessing.

λ11 λ21 L λm1 e11 e21 L ex1

λ λ22 M e12 e22 M
Λ(m, n) ∈ E ( x, y) = 12 ∈ (1)
M O M M O M
λ1n L M λmn e1 y L L exy

where Λ is the entire tooth area, and E is the tooth emphatic image in preprocessing. Third, the tooth
ratio of tooth width to tooth image was calculated using an oral structure composed of four incisors,
two cuspids, two premolars, and three molars on the upper and lower jaw. Figure 2 shows the calcula-
tion method and results. Then, teeth width per incisor, cuspid, premolar, and molar from each of the
other tooth images is acquired. Teeth ratio was acquired to compare the entire tooth area with tooth
size, and therefore, tooth sizes are assigned in the ratios of 5:3:2 per incisor, premolar, and molar, re-
spectively [7].
Finally, three tooth areas were made using the acquired tooth ratio. Figure 2(c) shows the results of
the division applied to the tooth ratio. Λ f is the incisor region, Λl and Λr are the side molar regions.
Eq. (2) shows the relation between the entire tooth area and the three divided areas.

Λ (m, n) = Λl (k , n) + Λ f ( j , n) + Λr ( z , n)
l
λ11 λl21 L λlk1 λ11f λ21f L λ fj1 r
λ11 λr21 L λrz1
l
λ12 λl22 M λ12f λ22f M r
λ12 λr22 M
= + +
M O M M O M M O M
f
λ1l n L L λlkn λ1n L L λ fjn λ1rn L L λrzn
(2)

(a) (b) (c)

Fig. 2. Trisection method using tooth morphological characteristic, (a) whole tooth area, (b) each tooth size, (c) division re-
sult.
3306 S.D. Na et al. / Individual tooth region segmentation using modified watershed algorithm

where n is the row of

o the image and k , j , an
nd z are the columns of the
t image.
The decrease in in
ntensity occu
urs in the sid
de molar regiions. Therefo
ore, enhancemment is perfo
formed li-
o Λl and Λr using each pixel location per regio
nearly on on and weigh ht α. Eq. (3)) shows the enhance-
ment meethod.

H = αΛl + Λ f + αΛr
l
λ11 λl21 L λlk1 λ11f λ 2f 1 L λ fj1 r
λ11 1 λr21 L λrz1
f f
⎛ 1 k ⎞λ
l
λl22 M λ λ M ⎛ r r
⎞ λ r
λr22 M (3)
= ⎜ ∑ + β ⎟ 12 + 12 22
+ ⎜ ∑ + β ⎟ 122
⎝ k =d d ⎠ M O M M O M ⎝ r =1 d ⎠ M O M
f f
λ1l n L L λlkn λ1n L L λ jn λ1rn L L λrzn

Where H is the enhaancement ressult, d is rang ons, and β iss the intensitty control
ge of the sidee molar regio
weight. Figure
F 3 shows the compparison beforre and after enhancement
e t.

2.3. Improved extracction method a seed usiing trisectionn areas

d of watershed boundary and

The watershed
w meethod as showwn in Figure 4 was based d on geograph hical methodd. The methood creates
pierced holds
h to form m the catchm
ment basins,, and catchm ment basins are
a formed ffrom pierced d holes to
dams, where
w a pierced hole and dam are seeed and bound dary in the watershed
w algorithm, respectively
[8,9]. Hoowever, in acccordance with
w the setup p for seed andd boundary in the convenntional waterrshed, the
dams an nd tooth borrders, respecttively. Consequently,
nd boundariees were consttructed at brright parts an
boundariies were seggmented betw ween tooth annd tooth. Hoowever, the method
m uses Sobel’s edge and the

Fig. 3. Enhancem ( image intensities befo

ment result, (a) ore and after enhancemennt, (b) prepro
ocessing
result and
a (c) applieed enhancem
ment result.
 S.D. Na et al. / Individual tooth region segmentation using modified watershed algorithm 3307

regional maximum value to address this disadvantage. In order to improve the problems of conven-
tional method, complement image is used on the watershed algorithm, as shown in Figure 4(b). The
extraction of boundary and seed is assignment method of different parameters at the each area. The
boundary of watershed was made to assign each thresholding per area. First, Otsu’s threshold method
to the incisor region is applied. Second, half of the incisor threshold to the premolar region is applied.
Finally, threshold value (25) to the molar region is applied. Then, each image was combined. Then,
the boundary of obtained combined image was extracted for the boundary of watershed. The seed is
acquired using the regional minimum value and morphology reconstruction. First, initial seed is ac-
quired using the regional minimum value on the tooth complement image. Then, reconstruction is per-
formed using the shrink and dilation morphology operation. Figures 4(c) and 4(d) show the resulting
boundary and seed, respectively.

3. Results

Figure 5 shows the results of the conventional and proposed methods. Figures 5(a) and 5(c) using
the conventional method reveal problems of non-detection in the molar region and over- and overlap-
segmentation of the incisors owing to Sobel’s edge characteristic and specular reflection. Figures 5(b)
and 5(c) utilizing the proposed method show improved tooth detection in the molar region, and in-
creased accuracy.
The accuracy of the proposed method was verified by applying it to each of the other images. Table
1 shows the comparison results of the conventional and proposed method. The proposed method de-
tects all the teeth in the molar regions and has no over-or overlap-segmentation due to specular reflec-
tion. Teeth number 4 and 5 show that the conventional method has problems in detecting tooth number
because of preprocessing that emphasizes the white parts of the tooth. Therefore, non-detections occur
owing to the difference from normal tooth color or a prosthetic tooth. Figure 6 shows that the proposed

(a) (b) (c) (d)

Fig. 4. Watershed algorithm principle of, (a) conventional watershed method, (b) complement watershed method, (c)
extracted boundary of watershed and (d) extracted seed of watershed.

(a) (b) (c)

Fig. 5. Tooth segmentation result, (a) original image, (b) conventional method segmentation and (c) proposed method
segmentation.
3308 S.D. Na et al. / Individual tooth region segmentation using modified watershed algorithm

method better prevents error rate than the conventional method. However, teeth numbers 3, 6, and 7
show partial over-segmentation due to specular reflection that is higher than that for other teeth or sim-
ilar entire tooth intensities.
In summary, Table 1 shows that the proposed method performs better than the conventional one in
terms of overall detection ratio and suppression of over- and overlap-segmentation.

4. Conclusion

The watershed algorithm is suitable to segment the tooth image, but has disadvantages of less accu-
racy in detecting tooth. In order to improve the problems, the trisection method was proposed, which
is a division method to tooth areas considering teeth width.
Our results show that the proposed method performs better than the conventional method. Therefore,
the proposed method was expected to simplify the inconvenience of prosthetic appliance manufacture
complexity and improve the location information efficiency for orthodontics by tooth shape.

Table 1
Results of tooth detection by conventional marker controlled watershed algorithm method and proposed method
MCWA method Proposed method
Error Error
Number
of teeth Detected
Detected tooth
tooth Over-seg. Overlap-seg. Over-seg. Overlap-seg.

Total 599 510 137 26 532 52 30

Average rate [%] - 86.39 28.85 5.28 90.3 10.6 5.14

50
MCWA

40 Proposed method
Error rate [%]

30

20

10

0
1 6 11 16 21
Images

Fig. 6. Error rate of MCWA and proposed method.

 S.D. Na et al. / Individual tooth region segmentation using modified watershed algorithm 3309

Acknowledgement

This research was financially supported by the Ministry of Knowledge Economy (MKE), Korea In-
stitute for Advancement of Technology (KIAT) and Dae-Gyeong Leading Industry Office through the
Leading Industry Development for Economic Region, and the “Over regional linked 3D convergence
industry promotion program” through the Ministry of Trade, Industry & Energy (MOTIE) and Korea
Institute for Advancement of Technology (KIAT).

References

[1] M. Rema and S.N. Madhu, Segmentation of human intestinal parasites from microscopy images using localized mean-
separation based active contour model, Biomedical Engineering Letters 3 (2013), 179–189.
[2] D. Hwang and G.L. Zeng, Special issue on medical imagen, Biomedical Engineering Letters 4 (2014), 1–2.
[3] G.-S. Lee and K. Park, Development of automatic scanning system for intra-oral scanner using SCARA robot, Proc. the
Society of CAD/CAM Conference, 2013, 332–337.
[4] F. Ullah, G.-S. Lee and K. Park, Development of a real-time 3D intraoral scanner based on fringe-projection technique,
Transactions of the Society of CAD/CAM Engineers 17 (2013), 156–163.
[5] Y.Q. Zhao, Q.P. Zhao and A.M. Hao, Multimodal medical image fusion using improved multi-channel PCNN, Bio-
medical Materials and Engineering 24 (2014), 221–228.
[6] S.-T. Lee, K.-S. Kim and T.-H. Yoon, Individual tooth image segmentation by watershed algorithm, The Trans. KIEE
59 (2009), 210–216.
[7] R.C. Sceid and G. Weiss, Woelfel's Dental Anatomy, Wolters Kluwer, Philadelphia, 2012, 110–143.
[8] G. Wood, Digital Image Processing Using MATLAB, Prentice-Hall, New Jersey, 2004, 567–636.
[9] Q. Wu, F.A. Merchant and K.R. Castleman, Microscope Image Processing, Academic Press, Massachusetts, 2008, 159–
194.