Implementation of Chain Code K-Nearest Neighbor in Qalqalah

Reading

Tjut Awaliyah Zuraiyah, Mulyati and Hanan Rosyada

Departement of Computer Science Pakuan University, Bogor, Indonesia

Keywords: Chain Code, Euclidean, KNN, Qalqalah.

Abstract: This study aims to implement the Chain Code method and KNN in Qalqalah reading law. The research phase

consists of image acquisition, pre-processing, grayscale, feature extraction using chain code, classification

using k-Nearest Neighbor k-NN). On the pre-processing process is carried out with image cropping, noise

removal and normalization image size 50x25 pixels. The data used are 132 datasets which are divided into 6

test data and 126 training data (63 original data and 63 rotation data). The Chain Code method is used as a

method of recognition of image patterns based on the direction of the wind, and the KNN method is used as

an image classification method based on the difference in euclidean distance. The KNN method will search

for the closest distance from the training data to the test data so that the best K is obtained with the highest

accuracy in each class. In the validation calculation, obtained the results of accuracy image classification with

an accuracy of 45.23%. This accuracy value is quite low because the Chain Code Method cannot detect images

with separate objects so that this method cannot work optimally.

1 INTRODUCTION

Al-Quran is a holy book that is owned by the Islamic

religion, which contains various kinds of life

meanings that can be applied in human life.

Al-Quran

provides an overview of life in the world and in the

hereafter and the effects that occur both positive and

negative impacts.

So, there must be tajweed learning

so that the message can be conveyed correctly.

Almost all Muslims in the world know how to read

the Quran, but some of them cannot read the Quran

correctly according to

makhraj and tajweed. People

who want to deep in the Quran will usually ask for

help from a teacher to learn more about the

makhraj

and tajweed recitation of the Quran correctly and

adequately.

A teacher will observe every makhraj

law and tajweed that are being read, then remind the

right tajwid law and guide the reader when there is an

error in reading.

This method is less effective because

a system is needed that can be used at any time when

needed.

At present, there are several Al-Quran

learning software on the market.

The usefulness of

this software is that users can only read the Quran, but

the placement of tajwid laws is not displayed.

This

software is still less effective for readers in learning

tajwid and messages contained therein.

Therefore, a

system is needed that can detect tajwid patterns so

that readers can find out how the laws of tajwid are

read, especially the recitation of qalqalah.

Based on

the related problems, a system was created entitled

Implementation of the Chain Code and k-NN

Method in the digital image-based recitation law of

qalqalah. Where chain code functions in pattern

detection while k-NN functions in the classification

of images obtained.

The tested image will be detected

based on the direction of the wind direction in the

chain code method.

2 RESEARCH METHOD

This study uses an image processing methodology

with several stages consisting of (1) the stage of

image acquisition, which

can be defined as a system

that functions to retrieve, collect and prepare data, to

process it to produce the desired data. (2)

Pre-

processing

stage is the stage to cut the character part

needed by the image. After the cutting results are

obtained, then the

noise removal is carried out (the

part that is not needed).

After that, the image size is

50 x 25 pixels uniform

. (3) The Grayscale stage is

244

Zuraiyah, T., Mulyati, . and Rosyada, H.

Implementation of Chain Code K-Nearest Neighbor in Qalqalah Reading.

DOI: 10.5220/0010623700002967

In Proceedings of the 4th International Conference of Vocational Higher Education (ICVHE 2019) - Empowering Human Capital Towards Sustainable 4.0 Industry, pages 244-250

ISBN: 978-989-758-530-2; ISSN: 2184-9870

changing the colour image into a

grey image scale,

this matter for simplify the image model. At this

stage, not only

Grayscale stages are used, but there

is also some other stage, namely as a binary that

works to change Grayscale

images to black and white

so that the image can be read clearly, namely

Imcomplement stage rebrand be the opposite of a

binary colour is white- black because usually, the

system can only read images with a black

background, then the Thinning

stage is the reduction

of the writing object so

that the obtained code chain

produces a short code sequence. (4) Chain Code

Method, this algorithm is formed with the aim of

representing the contours of an object.

The intended

representation includes pixels from the edges of an

object that are interconnected and have a certain

direction.

The workings of this algorithm to detect an

object is to give a rotation sign that is adjusted to the

direction of the wind that you want

to use, and (5)

Classification of KNN

is a pattern recognition stage

by comparing vector test data and vector training data

available using KNN.

Stage in this classification is to

determine the value of K, then calculate the euclidean

distance of each image for the next sort from smallest

to largest distance to get the best data by K.

Figure 1: Flowchart Research.

2.1 Image Acquisition

Data acquisition stage can be defined as a system that

functions to retrieve, collect and prepare data, to

process it to produce the desired data.

The resulting

image is not necessarily digital data, so it needs to be

digitized.

In this study, the tool used in image capture

is using a mobile camera. The following steps are

shown in figure 2:

Figure 2: Stage Acquisition

2.2 First Section

Pre-processing stage is the stage to perform the

required cutting portion character image.

After the

cutting results are obtained, then the noise removal is

done (the part that is not needed). After that, image

size uniformity is done.

Pre-processing stage,

including in Figure

Figure 3: Stage Preprocessing

This stage is shown in table 1 below:

Table 1: Pre-processing

2.3 Grayscale

The initial process many done in image processing is

to change the colour image to

grey image scale; this

is used for simplifying the image model.

At this stage, not only Grayscale stages are used,

but there is also some other stage, namely as

a binary

that works to change Grayscale images to black and

white so that the image can be read clearly, namely

Imcomplement stage rebrand be the opposite of a

Implementation of Chain Code K-Nearest Neighbor in Qalqalah Reading

245

binary colour is white-black because usually, the

system can only read images with a black

background, then the Thinning

stage is the reduction

of the writing object so that the obtained code chain

produces a short code sequence.

The following are the details of the stages:

Figure 4: Stage Grayscale

The example of the manual calculation is RGB image

changes to grayscale:

The total number of RGB

= 762052

Amount of Red Value = 248906

Amount of Green Value = 247324

Amount of Blue Value = 265822

R= 248906 = 0.326626

248906+247324+265822

G= 247324 = 0.32455

248906+247324+265822

B= 265822 = 0.348824

248906+247324+265822

Grayscale

RGB Value Search (2):

Grayscale = (0.2989*R)+(0.5870*G)+(0.1141*B)

= (0.2989*0.326626)+(0.5870*0.32455)

+(0.1141*0.348824)

= 0.32794

The following is an example in table 2:

Table 2: Grayscale

2.4 Metode Chain Code

The method used in this study uses the Chain Code

method. Algorithm this is formed with the aim of

representing the contours of an object.

The intended

representation includes pixels from the edges of an

object that are interconnected and have a certain

direction.

There are two types of Chain Code, namely

in part (a) using a four-way representation and part

(b) using an eight-way representation.

Each direction

shown has different values. The numbering scheme is

applied to this method.

The workings of this

algorithm to detect an object is to give a rotation sign

that is adjusted to the direction of the wind you want

to use.

Figure 5: Representation of the wind direction on the Chain

Code

The steps taken in the Chain Code are in figure 6:

Figure 6: Stage Chain Code

Examples of manual calculations for letter

detection in the Chain Code are:

1. Get internal contours

Figure 7: Results of Internal Contour Pattern

This pattern 1 on fig. 7 is a point pattern on qof ( ﻕ).

While on the right is the xy coordinator point that has

been obtained by inbound tracing.

Pattern 2 on fig.8 is qaf ( ﻕ) pattern without point.

While the right is the xy coordinator point obtained

by inbound tracing

ICVHE 2019 - The International Conference of Vocational Higher Education (ICVHE) “Empowering Human Capital Towards Sustainable

4.0 Industry”

246

Figure 8: Results of Internal Contour of Pattern 2

Figure 9: Results of Internal Contour Pattern 3

This pattern 3 on fig. 9 is the pattern of sukun

contained in the letter Qof ( ﻕ). While on the right is

the point of coordinator xy obtained.

2. Obtain Chain Code

The value of

Chain Code is obtained from the

detection of patterns based on the direction of the

wind

Figure 10: Chain Code Results

3. Obtain Image Size through the Chain Code

The area of the image can be done with the

Chain

code method, namely by the formula

Figure 11: Area of an object on a pattern

Figure 12. Area of Object in Pattern 2.

Figure 13: Area of Object in Pattern 3

Then the total area of all patterns is 99.5 + 12 + 16.5

= 128

2.5 Knn Method

The KNN Classification stage is a pattern recognition

stage by comparing the vector of test data and vector

of available training data using KNN.

The KNN

Classification Stage is shown in Figure

14 below:

Implementation of Chain Code K-Nearest Neighbor in Qalqalah Reading

247

Figure 14: KNN Method

In manual calculations, the data that has been

obtained previously will be calculated using the K-

NN method to determine the quality of the data that

is owned is good enough to be able to become training

data as a basis for making the program.

In the K-NN

method to determine the class seen from the data that

has the closest distance from the calculation test

method data.

To carry out the data visualization

process in this study, 132 datasets were used, which

consisted of 126 training data (63 original data and 63

rotation data) and 6 test data—the following

classification in Table

Table 3: Classification

The next step is to classify the data

testing by

calculating the square of the

euclid distance

(euclidean distance) of each object against the

training data provided. The Test Data Image is as

follows.

Table 4: Test Data and Training Data

Here's an example of the calculation:

Data = [ grayscale, area]

1. Test Data = [ 0.32794, 128]

2. Training Data 1 = [0.330697, 66]

3. Training Data 2 = [0.32956, 118]

4. Training Data 3 = [0.330695, 54.5]

Using formulas (3)







 X



X











(1)

Training Data 1







0.32794  0.330697²  128 66²

√

3844.000008

= 62.00000006

Training Data 2







0.32794  0.32956²  128  118²

√

100.0000026

= 10.00000013

Training Data 3









0.32794  0.330695









128  54.5





√

5402.250008

= 73.50000005

The Table

5 is the detected image value

Table 5. Detected Image Value

Trainin

Data Tes

Resul

Value

Trainin

Data 1 62.00000006

Trainin

Data 2 10.00000013

Trainin

Data 3 73.50000005

Calculation results from 3 training data on 1 test data

using

Grayscale results , Image Area and Euclidean

Distance based on Chain Code , test results obtained

from the comparison of the closest distance between

test data and training data, can be seen the smallest

value of the test result value is

10.00000013 owned

by Training Data 2 (reading

Qaf 2) , which is the

closest value to the test data, it can be concluded that

Training Data 2 (reading Qaf 2) is included in the

Test Data group (Read Qaf

1), where the two data

have in common namely entering into the

classification letter Qaf 1 .

3 RESULT AND ANALYSIS

In this study the Al-Quran image used was Al-Quran

Cordoba in Surah Al-Baqarah. In this image the Pre-

processing process is carried out by image cropping,

noise elimination and uniform image size 50x25

pixels. Next is the process of changing colored cira to

Grayscale, then Binary, Imcomplement and Thinning

. The data used are 132 datasets which are divided

into 6 test data and 126 training data (63 original data

ICVHE 2019 - The International Conference of Vocational Higher Education (ICVHE) “Empowering Human Capital Towards Sustainable

4.0 Industry”

248

and 63 rotation data). The Chain Code method is used

as a method of recognition of image patterns based on

the direction of the wind and the KNN method is used

as an image classification method based on the

difference in euclidean image distance. The KNN

method will search for the closest distance from the

training data to the test data so that the best K is

obtained with the highest accuracy in each class.

Figure 15: Display Pattern Detection using Chain Code

In Figure 15, it can be seen how the system processes

images

by detecting patterns using the Chain Code.

In this method, the image must be detected separately

in order to obtain accurate results.

This process starts

from image changes to grayscale,changes to

Grayscale to Binary, changes to binary to

Imcomplement ( negative from binary) and Thinning

(image reduction). After the image is processed, the

Chain Code value is searched to get the Area value

of the image.

Figure 16: Display of Image Classification Results

In figure 16, we can see the image classification

process in class 2 using the KNN method. This

method looks for image similarity by calculating the

closest distance between test data and training data

based on the grayscale value and the width of the

image

so that the best K value is obtained. In this

study, 6 Test data were used

and 126 training data (63

original data and 63 rotation data)

. Then obtained

must be the following accuracy:

Table 6: Test Results Image Classification Validation

Prediction

∑

Actual

Class

Actual

Class 1 2 0 0 0 0 1 3

Class 2 2 5 0 0 0 0 7

Class 3 0 0 3 0 0 0 3

Class 4 0 0 1 5 6 3 15

Class 5 0 0 2 2 1 2 7

Class 6 0 0 1 2 1 3 7

∑ Prediction 4 5 7 9 8 9 42

Table 7. Summary of Confusion Matrix Image

Classification

Prediction

∑

True False

Actual

True 19 23 42

False 0 0 0

∑ 19 23 42

Accuracy level for image classification test results:

𝐴𝑘𝑢𝑟𝑎𝑠𝑖 











*100% = 45.23%

𝑃𝑟𝑒𝑠𝑖𝑠𝑖 











*100% = 100%

𝑅𝑒𝑐𝑎𝑙𝑙 











*100% = 45.23%

𝐹𝑎𝑢𝑙𝑡 











*100% = 54.77%

4 CONCLUSIONS

Based on the results of the analysis of this study, it

can be concluded that the Chain Code Method can be

used to detect patterns in images and KNN can be

used in image classification by calculating the

Euclidean distance that has been sorted. In this study,

the best accuracy results in class 3 were obtained with

100% accuracy. This study uses an image processing

methodology with several stages consisting of the

stage of image acquisition, Pre-processing stage,

Grayscale stage, Chain- Method and Classification of

KNN. There is also a design stage consisting of

general design including the structure of the

navigation and system flowchart and detailed design

consisting of the design of the veranda, the design of

pattern detection and the design of image

classification. The implementation of this system

uses the Matlab R2011a application and calculations

in Microsoft Excel. This system has carried out trials

with several stages, namely structural testing,

functional testing and validation trials. All these trials

are done so that users can use them as best as possible

Implementation of Chain Code K-Nearest Neighbor in Qalqalah Reading

249

according to their needs. In the validation calculation,

obtained results of accuracy in image classification

obtained an accuracy of 45.23%. This accuracy value

is quite low because the Chain Code Method cannot

detect images with separate objects so that this

method cannot work optimally.

ACKNOWLEDGEMENTS

Thanks are given to Kemenristek Higher Education

which has provided support in the form of grants

Doctoral Research Grant; Foundation Pakuan

Siliwangi give Doctoral Scholarship and the

University Gunadarma as the study continued.

REFERENCES

Amani,Y.; Syahputra, I.; Siddiq, D. (2018). Sistem

Pendeteksi Pola Tajwid Al-Qur’an Hukum Mad thabi’I

Menggunakan Metode Sokal & Michener. TECHSI,

10(2), 15-29

Goel, V.; Jain, T.;Singhal S.; Kole S. (2017). Specific Color

Detection in Images using RGB Modelling in

MATLAB. International Journal of Computer

Applications, 161(8), 38-42.

Kartika, I. J.; Santoso E.; Sutrisno. (2017). Penentuan Siswa

berprestasi Menggunakan Metode K-Nearest Neighbor

dan Weight Product (Studi Kasus : SMP Negeri 3

Mejayan). Jurnal Pengembangan Teknologi Informasi

dan Ilmu Komputer, 1(5), 352-360.

Naufal, A. M. (2017). Implementasi Metode Klasifikasi K-

Nearest Neighbor (K-NN) untuk Pengenalan Pola Batik

Motif Lampung. Skripsi. Jurusan Ilmu Komputer,

Fakultas Matematika dan Ilmu Pengetahuan Alam,

Universitas Lampung, Lampung.

Santi, N. C. (2011). Mengubah Citra Berwarna Menjadi

Grayscale dan Citra Biner. Jurnal Teknologi Informasi

DINAMIK, 16(1), 14-19.

Septian, D. R. (2017). Sistem Deteksi Tajwid Al-Qur’an

menggunakan Metode Ekstraksi Fitur Canny dan SVM.

Skripsi. Jurusan Ilmu Komputer, Fakultas Matematika

dan Ilmu Pengetahuan Alam, Universitas Pakuan,

Bogor.

Siska, D.; Fadillah, C. (2016). Sistem Pendeteksi Pola

Tajwid Al-Qur’an Hukum Bi-Ghunnah dan Bila-

Ghunnah Pada Citra Menggunakan Metode Nei &

Li.Techsi, 8(1), 206-214

Wijaya, T. D. V.; Novianto, S.; Rosyidah, U. (2014).

Deteksi Huruf Arab Menggunakan Metode Freeman

Chain Code. Techno.COM, 13(4), 245-250.

Zuraiyah, T. A. (2016). Optimization of Feature Extraction

Using Combined Image Centroid Zone and Zone

Centroid Zone Method. Proceeding, International

Conference on Operation Research.

Virender Ranga Nikita Yadav, Pulkit Garg. (2018).

American Sign Language Fingerspelling Using Hybrid

Discrete Wavelet Transform-Gabor Filter And

Convolutional Neural Network. Journal of Engineering

Science and Technology, 13(9), 2655-2669.

Y. Boukhari, M. N.; Boucherit, M.; Zaabat, S.; Amzert, K;

Brahimi. (2018). Optimization Of Learning Algorithms

In The Prediction Of Pitting Corrosion. Journal of

Engineering Science and Technology, 13(5), 1153-

1164.

APPENDIX

Nomenclatures

N Amount of value

Tp True Positive

Fp False Positive

Fn False Negative

Euclidean Distance

test

Test Data

train

Training Data

Greek Symbols











Data Summary from k=1 until n

Abbreviations

KNN K- Nearest Neighbor

JESTEC Journal of Engineering Science and Technology

ICVHE 2019 - The International Conference of Vocational Higher Education (ICVHE) “Empowering Human Capital Towards Sustainable

4.0 Industry”

250