2020 3rd International Conference on Mechanical, Electronics, Computer, and Industrial Technology (MECnIT)

Harfu Jar Detection System In Al-Quran Using

Pierce Similarity Algorithm as a Basic Learning
Media of Arabic Language
Miftahul Jannah, Adli Abdillah Nababan
Software Engineering
STMIK Pelita Nusantara Medan, Indonesia
[email protected], [email protected]

Abstract— Arabic language has a very broad language using linear Vector Support Engine (SVM). There have been
structure, we need to know the most dominant Arabic many studies conducted in several Arabic cases, but there are
language found in the Al-Quran, so to know the interpretation no studies conducted related to the detection of Jar letters in
(Tafsir) of the Al-Quran, we must learn Arabic language. One the Quran using the image processing theory approach.
of the basic parts of Arabic language is nahwu science, nahwu therefore this research was built to detect the Jar letter.
is the study of laying row in Arabic such as the kasra, dhamma
and fatta. Jar letters is one of basic part in nahwu science. In The Recognition of the Jar letter can be done with a
this research, a detection system about Jar patterns will be preprocessing operation where the input data is an image of
developed with image processing approach. This system was the Quran. The algorithm used in this research is Pierce
built using Delphi XE with 7 sample of Jar letters where it will similarity. Where this method will read each pixel value of
be used in the training process. The process of detecting Jar the Al-Quran image to determine the level of similarity
patterns is using a method that will find the distance value between the training image vector pattern and the Testing
from training and testing process on the Al-Quran image. Image vector pattern. [9].
Training process stage begins with the bitmap extension file of
the original image,then change the size to gray scale level and II. METHODOLOGY
convolution edge detection so that it will produce a vector
This template, In this research, all research procedures must
value for each Jar pattern. Testing process will be use the
Pierce Similarity Algorithm to measure the distance value of be carried out like a literature study conducted by reading
the Jar pattern to be recognized. The percentage of system and gathering references related to the problem at hand. This
detection results obtained, the pierce similarity method is able study uses Borland Delphi in system development. The data
to recognize a Jar pattern of 60-80%. used in this study consisted of a collection of random Al-
Quran images. The data is obtained by scanning the sheets
Keywords— Image Processing; Pattern Recognition; Pierce of the Al-Quran then the file is saved in the bmp extension.
Similarity; Jar Letter.
A. Sample of Jar Pattern
I. INTRODUCTION (HEADING 1) The Jar letter training samples used in this study prove 7
Arabic is one of the international languages used in Jar letter images that represent the vector characteristics of
various countries. This language has a special character various Jar letters. The following Table shows some sample
pattern of writing and does not escape the prevailing sample Jars used as a Training process.
language order. One of the Arabic order is Nahwu Science,
where this science learns about determining the row in the 1) Training Sample
word or sentence. Jar letter In Arabic there are four types of The sample Jar letters used for training patterns are the Al-
row, first is fatta which is the top row in words or sentences, Quran imagery precisely Surah Shaad and Al-Anfal :
second is dhamma which are crooked row in words or
sentences, third is kasra which is the bottom line in words or
sentences and finally sukun which is a dead line in words or
sentences. As we know that we often find Arabic in the Al-
Quran . The Jar letter is important part of Nahwu science [1 -
2].
For some people to recognize a letter is an easy thing, but
this becomes a challenge for us to introduce to the machine. Fig.1. Sample of Jar pattern
Several studies conducted using several methods such as [3]
to recognize numbers in Arabic using Deep Learning Neural
Networks. [4] performed arabic handwriting recognition
using neural networks and get acuracy (98.50% for
characters, 96.90% for words). [5] conducting research on
sign language using hand drawing media. [6] conducting
research on the review of sign language recognition systems
on the basis of classification techniques using neural
networks. [7] conducting research to collect datasets in the
form of Arabic handwriting with two models namely,
connected handwriting and separate handwriting needs. [8]
conducting research to conduct research to classify images in
the form of Arabic handwriting that focuses on characters
that contain dots and characters without dots. classification

978-1-7281-7403-7/20/$31.00 ©2020
Authorized licensed use limited to: Heriot-Watt 349
IEEE Downloaded on
University. September 23,2020 at 12:41:09 UTC from IEEE Xplore. Restrictions apply.
 2020 3rd International Conference on Mechanical, Electronics, Computer, and Industrial Technology (MECnIT)

The image used in this study is an Al-Qur'an image

Surah Shaad Surah Al-Anfaal
containing a Jar letter pattern with image extensions
.bmp
2) Resizing Process
Resizing process is used to determine the pixel value
contained in the red box in the training sample, if
rezising is not there then automatically the pixel to be
recorded is the entire value in the image

Fig. 2. Sample of Jar pattern used In Al- Quran

The picture above shows that marked with a red box

shows a sample Jar letter that will be used for the training
process that produces vector values that will be used for
reference in the testing process
2) Testing Sample Fig. 4. Resizing Jar pattern Process
The testing process using 10 images of Al-Quran image of
Surah Hud, Al-A'raf, Yusuf, Al-maidah, Ghafir, As-Shaffat, At this stage, input Al-Quran is the original image
ibrahim, al-anam, At-taubah and Ar-ra'd. containing the Jar letter pattern will be resized in advance to
B. Scheme System save computation so that the calculation of the iteration of
each pixel coordinate value will be limited so that the
In this study, there were two stages carried out, namely pattern vector to be searched will be easier and more precise
the training process and the testing process. The stages in n this study the limits used are width = 600 and height =
this study began with the training process using original 400. Example of Resizing Process is Transform an Analog
images as input from the system being built. The inputted Image into a Digital Image and map letter patterns using a
data will undergo a resizing process to determine the size of shape as the following image:
the input image scale and then enter into the grayscale
process, convolution and Pierce Similarity methods. to
detect the pattern of Jar letters contained in the image of the
Al-Quran to determine the distance from the training and
testing of that image. The process can be illustrated in the
system scheme as shown Fig. 3:

Fig. 5. Example example of any pixel value

The picture above shows an example of the pixel value

Fig. 3. Scheme of system contained in an image that has been resized, so that images
that are not in the box will not be counted.
The details of the process of the system scheme are as
3) Grayscale Process
follows:
At this stage, Input Al-Quran is an image that
1) Image.bmp:
contains a Jar letter pattern that has been resized in
the previous process, the color will be changed to

350
Authorized licensed use limited to: Heriot-Watt University. Downloaded on September 23,2020 at 12:41:09 UTC from IEEE Xplore. Restrictions apply.
 2020 3rd International Conference on Mechanical, Electronics, Computer, and Industrial Technology (MECnIT)

grayscale. each component of the color Red, green then the image matrix will be multiplied by the sobel
and Blue will be added up and divided by the number operator.
of channels that is as many as three. Grayscale stage Sobel operator selection because it is more sensitive to
aims to equalize the intensity values of the three diagonal edges than vertical and horizontal edges. [13].
channels contained in the image.[11][12][15] Following is the form of the matrix of the sobel operator
a. Sbel horizontal operator: [16]

b. Sobel vertical operator :[16]

c. The sobel operator is the magnitude of the gradient [10]

[16]

TABLE I. EXAMPLE OF AN IMAGE PIXEL MATRIX

22 10 15 28
Fig. 6. Grayscale Process [14] 13 17 24 23
65 17 25 23
After resizing the image is done equalizing the intensity
into gray degrees then the next image will enter the
convolution stage. Table I shows the matrix of the pixels obtained in Fig 5
4) Convolution Process above, and then the calculation will be done using the Sobel
The picture below will illustrate the process of matrix
convolution Sx = (22)(-1) + (13)(-2) + (65)(-1) + 10(0) + 17(0) +17(0)
+(15)(1) + (24)(2) + (25)(1) = -25
Sy = (23)(1) + (10)(2) + (15)(1) + 13 (0 ) + 17(0) + 24(0) +
(65)(-1) + ( 17)(-2) + (25)(-1) = - 67
M= √((-25)^2+( (-67)^2 ) )= √((-25)^2 )+ √((-67)^2 )
= (-25)+(-62) = -92. [16].
Perform these calculations on all pixel values so that you
get the following results:

TABLE II. SOBEL IMAGE VALUE RESULTS IN MATRIX

-92 14 -32 -16 -4 -60

-96 22 -22 -12 26 -14
-4 20 8 -28 16 8

Table II contains the number of the M value that have been

calculated by the previous sobel operaror.

d. Pixel Normalization Process

Normalization = If M > 0 then M=1, else M=0

TABLE III. PIXEL NORMALIZATION VALUE RESULTS

0 1 0 0 0 0
0 1 0 0 1 0
Fig. 7. Convolution Process [14] 0 1 1 0 1 1

At this stage, the Al-Quran input is an image containing

a Jar letter pattern that has passed the grayscale process, Normalization is done to convert each pixel matrix into
numbers 1 and 0, where the value containing the object is
represented by 1 and the non-object value is represented by

351
Authorized licensed use limited to: Heriot-Watt University. Downloaded on September 23,2020 at 12:41:09 UTC from IEEE Xplore. Restrictions apply.
 2020 3rd International Conference on Mechanical, Electronics, Computer, and Industrial Technology (MECnIT)

0. after the normalization process, a vector pattern will be 1. Main Form

formed that represents each Jar letter character
5) Pierce Similarity Process
This process only applies to testing procedures. The
input vector is an image vector value that contains the
Jar letter pattern after the convolution process and
then performs the correction matrix calculation, and
ends with the weight matrix renewal. These steps are
repeated until a minimum error is reached or the
maximum iteration limit is met. [10]

Fig. 9. Main Form

Fig 9 shows the main Form is a form that becomes the

initial display of the Jar letter detection application.

2. Training Form

Fig. 8. Pierce Similarity Process

Where:
i= vector value of the training process image
j= vector value of the testing process image
a= the sum of the vector i , j which has the value i = 1, j=1
b= the sum of the vector i, j which has values i = 0, j=1
c= the sum of the vector i, j which has values i = 1, j=0
d= the sum of the vector i, j which has values i = 0, j=0

example of Pierce Similarity algorithm calculation

0 1 1 1 0 0 0 1 0 0
⎡0 1 0 1 0⎤ ⎡0 1 0 1 0⎤
⎢ ⎥ ⎢ ⎥
i= ⎢1 0 0 0 1⎥ 𝑗 = ⎢1 0 0 0 1⎥
⎢0 1 0 1 0⎥ ⎢0 1 0 1 0⎥
⎣0 1 1 1 0⎦ ⎣0 0 1 0 0⎦
𝑎 (1,1) = 8 𝑐 (1,0) = 4
𝑏 (0,1) = 0 𝑑 (0,0) = 13
𝑆
Fig. 10. Training Process Form

∗ ∗
𝑆𝑝 Fig 10 Shows The training form functions to get the Jar
∗ ∗ ∗ ∗ letter pattern vector used. The value of the vector pattern for
each Jar letter. Vector reference Jar pattern is related to the
𝑆𝑝 = =0 appearance of values 0 or 1 on a number of training Jar
letter images. The value 0 in the vector represents a value
III. RESULT AND DISCUSSION that does not belong to the features or characteristics of the
A. System Interface Jar letter but this value is within the Jar letter region,
The following is the appearance of the system being whereas the value 1 represents the value included in the
built: pattern features of the Jar letter.

352
Authorized licensed use limited to: Heriot-Watt University. Downloaded on September 23,2020 at 12:41:09 UTC from IEEE Xplore. Restrictions apply.
 2020 3rd International Conference on Mechanical, Electronics, Computer, and Industrial Technology (MECnIT)

measurement is done by training each Jar letter pattern.

Table IV explain the example of detecting the Jar letter
obtained at the Surah HUD.

TABLE IV. EXAMPLE OF JAR DETECTION IN SURAH HUD

(a) (b)

Fig. 11. (a) Vector Pattern of Jar Letters (b) enlarged image of the ‘ala
vector

Image (a) is a vector value representing each Jar letter

pattern, while image (b) is an enlarged image containing a
Jar letter vector, there is an image that the vector consists of
numbers 1 and 0 that have been obtained at the previous
normalization stage according in Table III above.

3. Testing Form

The Table IV shows an example of the detection carried

out in the testing sample at Surah Hud, where the variables
detected were 7 variables representing each Jar letter pattern
as ‘ala (J1), bi (J2), Hatta (J3), ‘An (J4), Fi (J5), Ila (J6) and
min (J7). Detection number shows how many letters the Jar
is able to detect in the Al-Quran being tested. Do the same
thing in 9 other sample testing as in the previous
methodology, So we get the following Table:

TABLE V. PERFORMANCE OF THE JAR DETECTION SYSTEM

Fig. 12 . Testing Process Form on Surah As-Shaffat

Fig 12 Shows The testing process is carried out for

detection of the image of the Al-Quran. This stage functions
as a result of the whole process.

B. System Performance "Number of Testing" shows the amount of testing data

Measurement performance evaluation system Jar letter available in the previous methodology, which is 10 suras for
detection generally uses two parameters, namely the True each Jar letter or variable. "Number of Image Detected"
Detection and false positive rate. Detection rate is the ratio shows how much the system is able to detect the Jar letter
or percentage of the number of Jar letters detected pattern of the total surah used. "Number of Image Not
successfully divided by the total number of Jar letters Detected" shows how much the system is unable to detect
tested, while the false positive rate is the number of of Jar the Jar letter pattern of the total surah used. The percentage
letters not detected successfully. System performance value obtained from the number of testing samples used is

353
Authorized licensed use limited to: Heriot-Watt University. Downloaded on September 23,2020 at 12:41:09 UTC from IEEE Xplore. Restrictions apply.
 2020 3rd International Conference on Mechanical, Electronics, Computer, and Industrial Technology (MECnIT)

as many as 10 Surah then divided by the number of images are grateful to our colleagues who provided insights and
detected then multiplied by 100. expertise that greatly assisted this research, even though they
may not agree with all interpretations and conclusions from
this paper.
REFERENCES
[1] A. S. Rahmat and M. Abdurrahman, “Analysis of Nahwu Learning
Progress in Indonesian Universities Based on the Character of Value
Concerns” in International Journal for Historical Studies, vol 8, no 2,
pp. 189-202,April. 2017.
[2] K. Park and D.Hyun Ki, “Accelerating Image Classification using
Feature Map Similarity in Convolutional Neural Networks”. Appl.
Sci, vol 9, no. 1, pp. 1-18, Nov. 2018.
[3] A.Ashiquzzaman and A. K.Tushar. “Applying Data Augmentation to
Handwritten Arabic Numeral Recognition Using Deep Learning
Neural Networks”.arXiv:1708.05969v4 [cs.CV], Sep. 2017
[4] A. Mars and G. Antoniadis. ”Arabic Online Handwriting Recognition
Using Neural Network” in International Journal of Artificial
Intelligence and Applications, vol. 7, no.5, pp. 51-59, Sep. 2016.
[5] R.Alzohairi, R. Alghonaim, W. Alshehri, etc. “Image based Arabic
Sign Language Recognition System” in International Journal of
Advanced Computer Science and Applications.(IJACSA), vol.9,no. 3,
Fig. 13 . System performance of Jar detection pp. 185-194. May. 2018.
[6] M. Mustafa. “A study on Arabic sign language recognition for
Fig. 13 Shows the graph illustrates the testing and differently abled using advanced machine learning classifiers”. In
measurement of performance performed after 7 vector Journal of Ambient Intelligence and Humanized Computing. March.
pattern Jar letters in the image are trained for the overall 2020.
characteristics of the image (J1, J2, J3, J4, J5, J6, J7). The [7] Y.Elarian, I. Ahmad, S. Awaidaa. “An Arabic Handwriting Synthesis
number of test images is 10 Al Quran images, for all stages System”. in Pattern Recognition, vol 48, no. 3, pp. 849-861. March.
2015.
of the training and the detection rate obtained for J1 = 80%,
[8] M. Anas Hussein A. “Offline Arabic Handwritten Isolated Character
J2 = 70%, J3 = 60%, J4 = 80%, J5 = 80%, J6 = 70%, J7 = Recognition System Using Support vector Machine and Neural
80%. Tthe performance of the detection rate system is Network” in Journal of Theoretical and Applied Information
strongly influenced by the source of the Jar letter pattern, Technology. vol. 95, no. 10. pp. 2315-2322. May. 2017.
observations of the reference pattern vector for testing, the [9] M. Bebah, C. Amine, M. Azzeddine and L. Abdelhak, ”Hybrid
Approaches For Automatic Vowelization Of Arabic Texts” in
complexity of the objects contained in the image, and the International Journal on Natural Language Computing (IJNLC) vol. 3,
success rate of the system in mapping the input Jar pattern no.4, pp. 53-71. August.2014.
vector into a vector Jar edge font pattern. [10] S.S. Choi, S.H. Cha, C. C. Tappert “A Survey of Binary Similarity
and Distance Measures”. in Journal of Systemics, Cybernetics and
IV. CONCLUSION Informatics, vol. 8, no. 1. pp. 43-48.2010.
The Conclusions of the research conducted, To get the [11] K. Anwar, A. Harjoko and S.Suharto, ”A New Method for Measuring
Texture Regularity based on the Intensity of the Pixels in Grayscale
Jar letter pattern in accordance with the sample being Images” in International Journal of Computer Applications, vol 137,
trained, the Jar letter pattern must be clear and not no. 7, pp. 0975-8887. March.2016.
connected between one letter and another. The lowest [12] K. Padmavathi and K. Thangadurai, ”Implementation of RGB and
percentage detection of Jar letter pattern is in hatta letter Grayscale Images in Plant Leaves Disease Detection–Comparative
Study”, in Indian Journal of Science and Technology, vol. 9, no. 6,
(J3) which is as much as 60%, this is because in the sample pp. 1-6. February. 2016.
testing not many contain the letter pattern while for the [13] S and J Munte, “Implementation of Sobel method to detect the seed
pattern 'ala (J1),' An (J4), Fi (J5) and min (J7) has the same rubber plant leaves” in Journal of Physics: Conference Series, vol.
percentage which is 80%, because of the Jar letter pattern 983, no. 1, pp. 1-5. March. 2018.
there are many in the sample testing. The similarity factor of [14] M,Jannah. “Facial expression system on video using widrow hoff”. In
each letter in the Koran becomes one of the weaknesses in Journal of Physics: Conference Series,vol. 978, pp. 1-5. March. 2018.
this system, because the system requires a threshold value [15] T. Kumar and K.Verma, “A Theory Based on Conversion of RGB
image to Grayimage” in International Journal of Computer
called the sensitivity value to overcome the problem of Application, vol. 7, no.2. pp. 0975–8887. September. 2010.
False detection patterns. [16] S. Achal and Shilpa J. “Analysis of Sobel Edge Detection Technique
for Face Recognition” in International Journal of Advanced Research
ACKNOWLEDGMENT in Computer Engineering & Technology (IJARCET), vol. 4, no. 5.
pp. 2450-2453. May. 2015.
This research was supported or partly supported by the
International Conference on Mechanics, Electronics,
Computers and Industrial Technology (MECnIT 2020). We

354
Authorized licensed use limited to: Heriot-Watt University. Downloaded on September 23,2020 at 12:41:09 UTC from IEEE Xplore. Restrictions apply.