Content uploaded by Agus Zainal Arifin
Author content
All content in this area was uploaded by Agus Zainal Arifin on Dec 29, 2014
Content may be subject to copyright.
Available via license: CC BY-NC-ND 4.0
Content may be subject to copyright.
TELKOMNIKA, Vol.10, No.1, March 2012, pp. 137~146
ISSN: 1693-6930
accredited by DGHE (DIKTI), Decree No: 51/Dikti/Kep/2010 137
Received August 15, 2011; Revised November 7, 2011; Accepted February 7, 2012
Classification and Numbering of Dental Radiographs
for an Automated Human Identification System
Anny Yuniarti*, Anindhita Sigit Nugroho, Bilqis Amaliah, Agus Zainal Arifin
Laboratory of Vision and Image Processing, Department of Informatics
Faculty of Information Technology, Institut Teknologi Sepuluh Nopember
Kampus ITS, Sukolilo, Surabaya 60111, Indonesia, Telp/fax 031-5939214/031-5913804
e-mail: anny@if.its.ac.id
Abstrak
Identifikasi manusia berbasis data gigi umum digunakan dalam forensik. Dalam kasus investigasi
yang besar, proses pengidentifikasian manusia yang dilakukan secara manual memerlukan waktu yang
lama. Pada makalah ini dikembangkan sebuah sistem identifikasi manusia otomatis menggunakan
radiografi gigi. Sistem yang dibangun bekerja dengan 2 tahapan utama. Tahapan pertama adalah
menyusun sebuah database berisi data radiografi gigi berlabel. Tahapan kedua adalah melakukan
pencarian pada database untuk mendapatkan hasil identifikasi. Kedua tahapan tersebut menggunakan
serangkaian proses pengolahan citra dan klasifikasi serta penomoran untuk mendapatkan pola dan fitur
radiografi gigi. Pertama, dilakukan prapemrosesan yang meliputi perbaikan dan binarisasi citra, ekstraksi
gigi tunggal, dan ekstraksi fitur. Selanjutnya, dilakukan proses klasifikasi gigi untuk mengklasifikasikan gigi
menjadi molar dan premolar dengan menggunakan metode binary support vector machine (SVM). Setelah
itu, proses penomoran pada gigi dilakukan sesuai pola molar dan premolar yang diperoleh pada tahap
sebelumnya. Percobaan menggunakan 16 radiografi gigi yang terdiri dari 6 radiografi bitewing dan 10
radiografi panoramik dengan total 119 objek gigi menunjukkan nilai akurasi yang baik, yaitu 91,6% untuk
proses klasifikasi gigi menjadi molar dan premolar dan 81,51% untuk proses penomoran gigi.
Kata kunci: forensik, identifikasi manusia, radiografi gigi, segmentasi, sistem penomoran gigi
Abstract
Dental based human identification is commonly used in forensic. In a case of large scale
investigation, manual identification needs a large amount of time. In this paper, we developed an
automated human identification system based on dental radiographs. The system developed has two main
stages. The first stage is to arrange a database consisting of labeled dental radiographs. The second
stage is the searching process in the database in order to retrieve the identification result. Both stages use
a number of image processing techniques, classification methods, and a numbering system in order to
generate dental radiograph’s features and patterns. The first technique is preprocessing which includes
image enhancement and binarization, single tooth extraction, and feature extraction. Next, we performed
dental classification process which aims to classify the extracted tooth into molar or premolar using the
binary support vector machine method. After that, a numbering process is executed in accordance with
molar and premolar pattern obtained in the previous process. Our experiments using 16 dental
radiographs that consist of 6 bitewing radiographs and 10 panoramic radiographs, 119 teeth objects in
total, has shown good performance of classification. The accuracy value of dental pattern classification
and dental numbering system are 91.6 % and 81.5% respectively.
Keywords: forensic, human identification, dental radiographs, segmentation, dental numbering system
1. Introduction
Biometric is a tool of identification that has been broadly used in many applications. A
biometric identification system is based on physical characteristics such as face [1], fingerprint,
palmprint [2], eyes (iris, retina) and DNA. However, many of those characteristics are only
suitable for ante mortem (AM) identification when a person to be identified is still alive. They
cannot be used for postmortem (PM) identification especially in the case of decay or severe
body damage caused by fire or collision [3].
Teeth are parts of human organ that are not easily decayed, located inside mouth and
thus they are more protected from decaying after human’s death. Therefore, teeth based
identification is one of reliable tools for PM identification.
TELKOMNIKA
Vol.10, No.1,
138
On average, human
has 32 teeth; each tooth has five surfaces, meaning that inside a
mouth there are 160 tooth surfaces with various conditions. If we use dental features as a tool
of identification, manual matching of AM with PM data needs a large amount of time and some
expertise. Therefore, a computer
In order to create an automated identification system, dental features on a dental
radiograph need to be extracted and saved in a database. During identification, features of e
tooth on the input are extracted and compared to those in the database. This matching process
will take a long time to complete if we do not reduce the search space. In this paper, we reduce
the search space by comparing the pattern and numbering of te
matched dental and numbering pattern. Therefore, we can enhance the effectiveness of the
identification system.
Figure 1 shows the international dental numbering system which also shows the molar
and premolar teeth. The
re are 32 teeth in adult people, sixteen teeth on each jaw. There are
two jaws, maxilla and mandible. Each jaw is divided into two groups, left and right. Thus, each
group consists of eight teeth comprised of two bicuspid, one cuspid, two premolar teeth, a
three molar teeth. In this research we only use molar and premolar teeth as part of dental
pattern, since molar and premolar teeth are usually stronger than others.
The international dental numbering system has teeth number from 1 to 32, starting from
t
he third molar in the right maxilla (#1), going through the maxilla to the third molar in the left
maxilla (#16). Next, the numbering is continued to the third molar in the left mandible (#17) and
around the mandible until we find the third mola
There are three kinds of dental radiographs: bitewing, panoramic, and periapical.
literatures [3 - 5
], bitewing images are usually used for identification. However, in this paper, we
tested our method not only to bitewing radiogr
radiographs have wider space between upper and lower jaw, whereas, in panoramic
radiographs, the upper and lower jaw are closer.
Automated dental based identification consists of extracting dental feature
matching itself [5 - 7
]. In this paper, the dental feature used for identifying human is the
arrangement of Molar and Premolar teeth and the numbering of each radiograph. In order to
have this arrangement, we have to classify each tooth in a r
But first, we have to extract the tooth using several image processing techniques. The tooth
separation is crucial to the system. Our tooth separation method has been able to correctly
extract single tooth. There are only
segmented due to very high intensity in the lower jaw bone.
The rest of the paper is organized as follows. Section 2 gives an explanation of the
method used in this research. Section 3 explains the result
present the conclusion and future works.
1
2
32
31
Right
Figure 1. A system of dental numbering in adults
2. Research Method
In this section, the proposed system design and three main functions, namely pre
processing, teeth separation, classification and numbering system, are explained. All functions
in the proposed system are implemented using Matlab 7.0.
There are two main pha
Figure 2. They are dental data recording phase and identification phase. In the dental data
recording phase, dental radiographs are processed. There are three main functions in this
phase, namely
preprocessing, teeth separation, classification and numbering. The results of this
phase are dental patterns, which next to be recorded in a database along with the original
Vol.10, No.1,
March 2012 : 137 – 146
has 32 teeth; each tooth has five surfaces, meaning that inside a
mouth there are 160 tooth surfaces with various conditions. If we use dental features as a tool
of identification, manual matching of AM with PM data needs a large amount of time and some
expertise. Therefore, a computer
-
aided for an identification system is needed.
In order to create an automated identification system, dental features on a dental
radiograph need to be extracted and saved in a database. During identification, features of e
tooth on the input are extracted and compared to those in the database. This matching process
will take a long time to complete if we do not reduce the search space. In this paper, we reduce
the search space by comparing the pattern and numbering of te
eth only. This results in a list of
matched dental and numbering pattern. Therefore, we can enhance the effectiveness of the
Figure 1 shows the international dental numbering system which also shows the molar
re are 32 teeth in adult people, sixteen teeth on each jaw. There are
two jaws, maxilla and mandible. Each jaw is divided into two groups, left and right. Thus, each
group consists of eight teeth comprised of two bicuspid, one cuspid, two premolar teeth, a
three molar teeth. In this research we only use molar and premolar teeth as part of dental
pattern, since molar and premolar teeth are usually stronger than others.
The international dental numbering system has teeth number from 1 to 32, starting from
he third molar in the right maxilla (#1), going through the maxilla to the third molar in the left
maxilla (#16). Next, the numbering is continued to the third molar in the left mandible (#17) and
around the mandible until we find the third mola
r in the right mandible (#32) [4
].
There are three kinds of dental radiographs: bitewing, panoramic, and periapical.
], bitewing images are usually used for identification. However, in this paper, we
tested our method not only to bitewing radiogr
aphs, but also to panoramic radiographs. Bitewing
radiographs have wider space between upper and lower jaw, whereas, in panoramic
radiographs, the upper and lower jaw are closer.
Automated dental based identification consists of extracting dental feature
]. In this paper, the dental feature used for identifying human is the
arrangement of Molar and Premolar teeth and the numbering of each radiograph. In order to
have this arrangement, we have to classify each tooth in a r
adiograph into Molar or Premolar.
But first, we have to extract the tooth using several image processing techniques. The tooth
separation is crucial to the system. Our tooth separation method has been able to correctly
extract single tooth. There are only
two of sixteen images that have not been correctly
segmented due to very high intensity in the lower jaw bone.
The rest of the paper is organized as follows. Section 2 gives an explanation of the
method used in this research. Section 3 explains the result
s and analysis. In Section 4, we
present the conclusion and future works.
Right Maxilla
Left Maxilla
3
4
5
6
7
8
9
10
11
12
13
14
15
16
30
29
28
27
26
25
24
23
22
21
20
19
18
17
Right
Mandible
Left Mandible
Figure 1. A system of dental numbering in adults
In this section, the proposed system design and three main functions, namely pre
processing, teeth separation, classification and numbering system, are explained. All functions
in the proposed system are implemented using Matlab 7.0.
There are two main pha
ses in the proposed human identification system as shown in
Figure 2. They are dental data recording phase and identification phase. In the dental data
recording phase, dental radiographs are processed. There are three main functions in this
preprocessing, teeth separation, classification and numbering. The results of this
phase are dental patterns, which next to be recorded in a database along with the original
ISSN: 1693-6930
has 32 teeth; each tooth has five surfaces, meaning that inside a
mouth there are 160 tooth surfaces with various conditions. If we use dental features as a tool
of identification, manual matching of AM with PM data needs a large amount of time and some
aided for an identification system is needed.
In order to create an automated identification system, dental features on a dental
radiograph need to be extracted and saved in a database. During identification, features of e
ach
tooth on the input are extracted and compared to those in the database. This matching process
will take a long time to complete if we do not reduce the search space. In this paper, we reduce
eth only. This results in a list of
matched dental and numbering pattern. Therefore, we can enhance the effectiveness of the
Figure 1 shows the international dental numbering system which also shows the molar
re are 32 teeth in adult people, sixteen teeth on each jaw. There are
two jaws, maxilla and mandible. Each jaw is divided into two groups, left and right. Thus, each
group consists of eight teeth comprised of two bicuspid, one cuspid, two premolar teeth, a
nd
three molar teeth. In this research we only use molar and premolar teeth as part of dental
The international dental numbering system has teeth number from 1 to 32, starting from
he third molar in the right maxilla (#1), going through the maxilla to the third molar in the left
maxilla (#16). Next, the numbering is continued to the third molar in the left mandible (#17) and
].
There are three kinds of dental radiographs: bitewing, panoramic, and periapical.
In
], bitewing images are usually used for identification. However, in this paper, we
aphs, but also to panoramic radiographs. Bitewing
radiographs have wider space between upper and lower jaw, whereas, in panoramic
Automated dental based identification consists of extracting dental feature
s and feature
]. In this paper, the dental feature used for identifying human is the
arrangement of Molar and Premolar teeth and the numbering of each radiograph. In order to
adiograph into Molar or Premolar.
But first, we have to extract the tooth using several image processing techniques. The tooth
separation is crucial to the system. Our tooth separation method has been able to correctly
two of sixteen images that have not been correctly
The rest of the paper is organized as follows. Section 2 gives an explanation of the
s and analysis. In Section 4, we
In this section, the proposed system design and three main functions, namely pre
-
processing, teeth separation, classification and numbering system, are explained. All functions
ses in the proposed human identification system as shown in
Figure 2. They are dental data recording phase and identification phase. In the dental data
recording phase, dental radiographs are processed. There are three main functions in this
preprocessing, teeth separation, classification and numbering. The results of this
phase are dental patterns, which next to be recorded in a database along with the original
TELKOMNIKA ISSN: 1693-6930
Classification and Numbering of Dental Radiographs for An Automated …. (Anny Yuniarti)
139
radiographs. The identification phase aims to identify a dental radiograph, called a query,
belongs to which data in the database. The functions applied to the dental radiograph in the
identification phase are similar to those applied to radiographs in the recording phase. The
result of classification and numbering system in this phase is used as a search query which
leads to an identification result.
Figure 2. Design of the proposed human identification system
In the pre-processing step, a dental radiograph which has been digitalized are loaded
from local hard disk. The dental radiograph can be bitewing or panoramic. For panoramic
radiographs, we only take the molar and premolar part. Next, we perform image enhancement
that aims to equalize the brightness level so that there is no pixel that has very high intensity
level compare to its neighbors. This usually happens to dental fillings. Next, we perform the
contrast enhancement. Generally, dental radiographs have low contrast. In order to facilitate
process of teeth separation with the background, we increase the contrast using morphological
operation and top-hat and bottom-hat operator [8]. After that, we perform local histogram
equalization which is called Contrast-Limited Adaptive Histogram Equalization (CLAHE) [8].
After pre-processing, the grayscale digital radiographs are then converted into binary
images using Otsu's thresholding method [8] followed by closing and opening operation to
smooth the teeth contour and remove noises. Next, we perform horizontal integral projection [9]
followed by spline method to separate the image into a maxilla image and a mandible image.
Finally, we use the vertical integral projection method on each maxilla and mandible image
independently to extract single tooth image.
The next process is dental feature extraction of each tooth. This step is used for
classifying each tooth into molar or premolar class. The dental features are area of each tooth
and ratio of each tooth's width and height. After feature extraction, we classify each tooth into
molar or premolar class using binary support vector machine (SVM) method. SVM is a famous
binary classification method. Given a set of training data, each marked as belonging to one of
two classes, an SVM training method creates a model that predicts a new data is in one class or
the other [10]. An SVM model is a representation of all data as points in space and a clear gap
that separate data into two categories. This clear gap is often called as a hyperplane. This
hyperplane is built as wide as possible. New testing data are then mapped into the same space
and predicted as a member of a class based on which side of the hyperplane they fall on.
The molar-premolar pattern of each image is then refined using default patterns. There
are two kinds of default patterns, as shown in Figure 3, namely patterns for right side of teeth
(Pattern 1) and patterns for left side (Pattern 2). In this paper, the first step of numbering system
is find which default pattern that has highest similarity value with the pattern sequence tested.
The similarity matrix is computed using simplified Smith-Waterman algorithm [11] as in Equation
(1). Let
m
tttT K
21
=
be a sequence of dental numbering,
m
pppP K
21
=
be a dental pattern and
nm
≤
. The similarity matrix O = {O
ij
} consists of similarity degrees between T
i
and P
j
segment
pair.
Dental data recording phase
Identification phase
Dental
Database
Search
Query
Dental
Radiographs Pre-
processing Teeth
Separation
Classification
and
Numbering
A Dental
Radiograph Pre-
processing Teeth
Separation
Classification
and
Numbering
Identification
Result
ISSN: 1693-6930
TELKOMNIKA Vol.10, No.1, March 2012 : 137 – 146
140
≠− =+
=
−−
−−
jiji
jiji
ij
ptkO
ptO
O if}0,,max{
if1
3
1
1,1
1,1
(1)
Using Equation (1), we compute four similarity matrices, namely Omax
1
, Omax
2
,
Oman
1
, and Oman
2
. Then the maximum value of Omax
1
is added to that of Oman
1
and
compared to sum of the maximum value of Omax
2
and Oman
2
. If the sum of the maximum value
of Omax
1
and Oman
1
is higher than that of Omax
2
and Oman
2
, then we choose Pattern 1.
Otherwise, we choose Pattern 2. The next step is defining position of dental numbering based
on the chosen default pattern. First, find an element of similarity matrix O
kl
that has maximum
value, set a column index and a row index based on the element’s position, i.e. the column
index = k, and the row index = l. Then, label each tooth in maxilla and mandible with number as
in default pattern numbering system, i.e. p
l-k+i
, 1 ≤ i ≤ k.
As an illustration, suppose that patterns resulted from the SVM classification results are
molar-molar-molar-premolar-premolar (MMMPP) for maxilla and MMPP for mandible. Using
equation (1), the similarity matrices are as shown in Figure 4. Then find the maximum value of
each matrix, i.e. Smax
1
=5, Smax
2
=3, Sman
1
=4, Sman
2
=2. Thus, the score of Pattern 1 is 9,
while the score of Pattern 2 is 5. Therefore we choose Pattern 1 as the default pattern. Next,
label each tooth using teeth alignment method described above. For the maxilla sequence
(MMMPP), k = 5 and l = 5. Then the resulted number sequence is 1-2-3-4-5. For the mandible
sequence (MMPP), k = 4 and l = 5. Therefore the number sequence is 31-30-29-28.
Pattern 1
M
M
M
P P P
Maxilla 1 2 3 4 5 6
Mandible
32
31
30
29
28
27
Pattern 2
P P P M
M
M
Maxilla 11
12
13
14
15
16
Mandible
22
21
20
19
18
17
Figure 3. Two default patterns of dental numbering
Figure 4. The similarity matrices between two default patterns and the pattern MMMPP-MMPP.
The last procedure in the proposed identification system is related to database access.
Final images after the classification and numbering process are stored in the database along
with dental data such as a unique serial number, name and age of the radiograph’s owner, date
or recording, molar-premolar pattern, numbering pattern, area and ratio features, and file path of
the original image, picture of the owner, and the classified image.
We use the pattern of molar-premolar and numbering both in the maxilla and mandible
as the query of identification process. The result of this kind of query may include more than one
identified person. For further processing, user may add area and ratio features as part of the
query. Using these features, the system will choose data in the database that has equal area
and ratio.
TELKOMNIKA ISSN: 1693-6930
Classification and Numbering of Dental Radiographs for An Automated …. (Anny Yuniarti)
141
3. Results and Discussion
We use 16 dental radiograph images, composed of 6 bitewing radiographs and 10
panoramic radiographs. Based on an expert identification, there are 37 teeth objects identified
in the 6 bitewing radiographs. Whereas in the panoramic radiographs, there are 82 teeth objects
identified by the expert. Therefore, there are 119 objects of tooth in total. Three samples of the
system's input image are as shown in Figure 5(a-c).
3.1. Pre-processing and Teeth Separation
In the first process, input images are successfully enhanced as shown in Figure 6.
However, in the case of tooth object having too low intensity approaches background's intensity,
or in the case of lower jaw bone having too high intensity approaches teeth object's intensity,
this process does not perform well, as in three out of sixteen images in our experiment.
In the binarization process, the enhanced images are successfully converted into binary
images. Except for the three images having the intensity problem as we explained before, all
binary images have the properties as follows. The white pixels of the binary images represent
teeth objects, whereas non-hole black pixels represent background. Sample outputs of the
binarization process are as shown in Figure 7.
(a) (b) (c)
Figure 5. Sample input to the system: (a) a bitewing radiograph (b) a left-cropped panoramic
radiograph (c) right-cropped panoramic radiograph
(a) (b) (c)
Figure 6. Results of enhancement process applied to three sample inputs as in Figure 5.
(a) (b) (c)
Figure 7. Results of binarization process applied to three enhanced images as in Figure 6(a-c).
The process after binarization is separating each radiograph into two parts, namely
maxilla and mandible part. Our experiments show that using the horizontal integral projection
followed by spline method, we can split the radiograph into two regions (maxilla and mandible)
well (see Figure 8). Figure 8(a) and 8(d) are the two regions resulted from Figure 7(a). We can
see that bitewing radiographs are easier to be horizontally separated. Figure 8(b, e) and 8(c, f)
ISSN: 1693-6930
TELKOMNIKA Vol.10, No.1, March 2012 : 137 – 146
142
are resulted from Figure 7(b) and (c) respectively. Here, we can successfully separate the
binary image into maxilla and mandible images although the upper and lower jaws are very
close in panoramic radiographs.
Each region is processed further by applying the vertical integral projection followed by
spline method to separate the teeth region into single tooth region. Overall our method performs
well in our experiments except for two images that have very high lower jaw bone intensity. In
the case of molar tooth that has double roots in mandible, our method performs well too
because we only take 3/5 upper part of mandible. Hence pixel values of tooth root are not
included in the computation of teeth separation.
3.2. Classification and Numbering
In the classification process, firstly we considered a tooth object is an isolated area
having more than 6000 pixels. From each tooth object, we extracted its area, ratio of height and
width, and its centroid. Based on these features, we classify each tooth into molar or premolar
using binary SVM method. As a comparison, we also implemented the classification using k-
nearest neighbor (kNN) method, a simpler method than SVM, with k = 9.
Based on our experiments, there is significant difference between accuracy of SVM
classification result and that of kNN’s result. Using the SVM method, the total accuracy value
reaches 89.07% or 106 out of 119 objects were truly classified. Whereas, the average accuracy
of the kNN method reaches 77.31% or 92 out of 119 objects were truly classified.
Next, we applied the numbering system by marking all teeth using a number and we
also modified the class using standard numbering system in order to avoid abnormal molar and
premolar pattern. As an example, if a classification process results in a pattern such as
premolar-molar-premolar-premolar (P-M-P-P), then the pattern will be modified into M-M-P-P.
This strategy has been able to improve the system’s accuracy to 91.60%. Hence, 109 out of 119
objects are now classified correctly. The implemented numbering system also performs well.
There are 97 out of 119 objects numbered correctly. This leads to a total accuracy value of
81.51%. Details of classification and numbering accuracy value are shown in Table 1 and Table
2 respectively. Whereas, sample output images are shown in Figure 9. In Figure 9, extracted
teeth are marked using a yellow line, labeled by M for molar class or P for premolar class
followed by a number representing the numbering's result.
(a) (b) (c)
(d) (e) (f)
Figure 8. Results of teeth separation applied to three binary images as in Figure 7(a-c);
Top row: maxilla regions. Bottom row: mandible regions.
(a) (b) (c)
Figure 9. Results of classification and numbering process applied to extracted teeth as
in Figure 8(a-f).
Table 1. The accuracy of molar-premolar classification
TELKOMNIKA ISSN: 1693-6930
Classification and Numbering of Dental Radiographs for An Automated …. (Anny Yuniarti)
143
No Filename
Classification Accuracy (%)
kNN SVM SVM followed
by default
pattern
modification
1 bit1_Right.tif 100.00 100.00 100.00
2 bit2_Right.tif 71.42 100.00 100.00
3 bit3_Left.tif 80.00 80.00 80.00
4 bit4_Right.tif 71.42 100.00 100.00
5 bit5_Left.tif 57.14 71.42 85.71
6 bit6_Right.tif 33.33 66.67 66.67
7 pan1_Left.tif 100.00 100.00 100.00
8 pan1_Right.tif 87.50 100.00 100.00
9 pan25_Left.tif 88.89 88.89 88.89
10 pan25_Right.tif 100.00 100.00 100.00
11 pan34_Left.tif 75.00 100.00 100.00
12 pan34_Right.tif 71.42 100.00 100.00
13 pan50_Left.tif 50.00 75.00 75.00
14 pan50_Right.tif 55.56 55.56 66.67
15 pan70_Left.tif 87.50 100.00 100.00
16 pan70_Right.tif 100.00 87.50 100.00
Total accuracy out of 119 tooth
objects 77.31 89.07 91.60
Table 2. The accuracy of numbering using teeth alignment
No Filename Numbering Accuracy (%)
1 bit1_Right.tif 60.00
2 bit2_Right.tif 100.00
3 bit3_Left.tif 60.00
4 bit4_Right.tif 100.00
5 bit5_Left.tif 85.71
6 bit6_Right.tif 00.00
7 pan1_Left.tif 100.00
8 pan1_Right.tif 100.00
9 pan25_Left.tif 88.89
10 pan25_Right.tif 100.00
11 pan34_Left.tif 100.00
12 pan34_Right.tif 100.00
13 pan50_Left.tif 50.00
14 pan50_Right.tif 33.67
15 pan70_Left.tif 100.00
16 pan70_Right.tif 100.00
Total accuracy out of 119 tooth
objects 81.51
3.3. Identification System
The proposed automated human identification system was implemented using MySQL
database server and Matlab 7.0. The system consists of four user interfaces. The first user
interface is used for classification and numbering of dental radiographs. Sample input and
output of the classification and numbering system is as shown in Figure 10. In the system's user
interface, there are 6 buttons consisting of "Open Image" button to load an input image from
local disk, "Proceed" button to perform the proposed methods, "Save" button to store the
radiograph and its properties including dental pattern and numbering into the database,
"Search" button to find a match of current radiograph in the database based on its properties,
"Database" button to browse the database's contents, and "Exit" button to quit the application.
The second user interface aims to add a classified radiograph into the database. This
interface only appears after users click the “Save” button in the first user interface. In this
window, users may add additional information such as name, age, picture, and other
information. Figure 11 shows an example of the action.
The third window appears when users click the “Search” button in the first user
interface. This window aims to find whether there is matched data in the database based on the
resulted pattern and numbering. The search process may results in zero, one, or more than one
identity. This is because we only comparing the dental pattern including the dental numbering.
Figure 12 shows an example when the system found exactly one matched result.
ISSN: 1693-6930
TELKOMNIKA Vol.10, No.1, March 2012 : 137 – 146
144
Figure 10. The classification and numbering system's user
interface.
Figure 11. The user interface for
saving new data into database.
Figure 12. The user interface for searching process. Right image is the query, left image is the
result.
The last user interface aims for viewing and querying the database. In this window,
users are able to view all data in the database and to execute query based on pattern or
numbering. Figure 13 illustrates query “14-15-16” that results in one found data.
TELKOMNIKA ISSN: 1693-6930
Classification and Numbering of Dental Radiographs for An Automated …. (Anny Yuniarti)
145
Figure 13. The user interface for querying database. Users are asked to enter the pattern or
numbering in the Search textfield.
4. Conclusion
The proposed system has been successfully implemented and is able to generate
dental pattern and numbering based on dental radiographs. In this paper, we have shown that
our method can be applied not only to bitewing radiographs, but also to panoramic radiographs.
The total accuracy value of dental pattern classification is 91.6% and the total accuracy of
dental numbering system is 81.5%. However, there are some images that cannot be segmented
correctly, due to low intensities of tooth objects. This error propagates into next processes and
hence leads to incorrect classification and numbering. Therefore, the segmentation method still
needs further research.
Acknowledgment
The authors wish to acknowledge the Institute of Research and Public Services, Institut
Teknologi Sepuluh Nopember (ITS), which has financed the program through the letter of
agreement implementation research: 781/I2.7/PM/2011 Date: 1 April 2011. Dental radiographs
used in this research are obtained from the Hiroshima University Hospital, Hiroshima, Japan.
References
[1] Muntasa A, Sirajudin IA, Purnomo MH. Appearance Global and Local Structure Fusion for Face Image
Recognition. TELKOMNIKA Indonesian Journal of Electrical Engineering. 2011; 9(1): 125-132.
[2] Putra IKGD, Erdiawan. High Performance Palmprint Identification System Based On Two Dimensional
Gabor. TELKOMNIKA Indonesian Journal of Electrical Engineering. 2010; 8(3): 309-318.
[3] Lin PL, Lai YH, Huang PW. An Effective Classification and Numbering System for Dental Bitewing
Radiographs using Teeth Region and Contour Information. Pattern Recognition Journal. 2010; 43:
1380-1392.
[4] Mahoor MH, Abdel-Mottaleb M. Classification and Numbering of Teeth in Dental Bitewing Images.
Pattern Recognition Journal. 2005; 38: 577-586.
[5] Abdel-Mottaleb M., Nomir O, Nasser DE, Fahmy G, Ammar H. Challenges of Developing an
Automated Dental Identification System. 64th IEEE Midwest Symposium on Circuits and Systems.
Cairo, Egypt. 2004.
ISSN: 1693-6930
TELKOMNIKA Vol.10, No.1, March 2012 : 137 – 146
146
[6] Ammar H, Abdel-Mottaleb M, Jain A. Automated Dental Identification System (ADIS). 8th Annual
International Conference on Digital Government Research: Bridging Discipline & Domains.
Philadelphia, Pennsylvania, USA. 2007; 228: 248-249.
[7] Samopa F. Tooth Shape Measurement on Dental Radiographs for Forensic Personal Identification.
Dissertation of Department of Information Engineering, Graduate School of Engineering, Hiroshima
University. Hiroshima, Japan; 2009.
[8] Gonzalez RC, Woods RE. Digital Image Processing Using MATLAB. New Jersey, USA: Pearson
Prentice-Hall, Pearson Education, Inc. 2004.
[9] Hermawati FA, Koesdijarto R. A Real-Time License Plate Detection System for Parking Access.
TELKOMNIKA Indonesian Journal of Electrical Engineering. 2010; 8(2): 97-106.
[10] Duda RO, Hart PE, Stork DG. Pattern Classification Second Edition. New York, USA: Wiley-
Interscience. 2001.
[11] Smith TF, Waterman MS. Identification of common molecular subsequences. Journal of Molecular
Biology. 1981; 147: 195–197.