Batik Classification using Texture Analysis and Multiclass Support

Vector Machine

Wahyu Tri Puspitasari, Dian C. R. Novitasari and Wika Dianita Utami

Departement of Mathematics UIN Sunan Ample Surabaya, Ahmad Yani Street, Surabaya, Indonesia

Keywords: Batik Image, Analysis Texture, Feature Extraction, GLCM, DWT Multiclass SVM.

Abstract: Batik is one of the cultural heritage has become an Indonesian identity and recognized by the Organization of

Education, Science and Cultural Organization (UNESCO). Every region in Indonesia has very diverse batik

motifs. There are 38 batik motifs based on the area of origin. It will be difficult to recognize each of these

patterns while batik began to be liked by many local and foreign tourists. Therefore, a system is needed that

can recognize every pattern of batik to facilitate people in recognizing batik motifs. Support Vector Machine

(SVM) has excellent performance in classification and can also be used to recognize patterns of batik motif.

We use the Gray Level Co-occurrence Matrix (GLCM) for feature extraction and SVM for batik classification.

The result show that batik motif can be classified using SVM with 96% accuracy for two types of batik motifs,

88.89% for three types of batik motifs and 77.14% for four types of batik motifs.

1 INTRODUCTION

Batik is Indonesia’s cultural heritage that has been

worldwide. Batik is a fabric with patterns, motifs, or

certain themed images according to the philosophy

that exist in the regions of Indonesia (Wulandari,

2012). Batik motif in Indonesia is very diverse

because the Indonesian nation has a diversity of

ethnic and cultures. Motif consists of elements

proportion and composition. Indonesia has more than

181 batik motifs (Achjadi, 1999). The types of batik

can be classified by method of manufacture, area of

origin and motif.

(a) (b) (c)

Figure 1: (a) Lereng Madura batik image, (b) Ceplok

Indramayu batik image, (c) Sidomukti Yogyakarta batik

Image.

Some examples of batik motifs in Figure 1 are

batik motifs based on area of origin. Batik motifs in

Indonesia are difficult to differentiate because the

amount are very diverse. To recognize batik motifs, a

classification process is needed. Classification is used

to recognize the characteristics of the objects

contained in the database and classed into different

class (Moertini & Sitohang, 2005). The process of

batik classification is the division of the image of

batik into the classes in accordance with the pattern

of the motive, it makes more easily recognizable

based on the pattern.

Texture classification can be done using texture

analysis results. Texture analysis in the image is an

observation about a characteristic in the image. To get

the characteristics of the image be done by extracting

the image feature that serves to take the features of

each image. The feature extraction methods include

Cardinal Spline Curve Representation (Fanani,

Yuniarti, & Suciati, 2014), FPGA (Babasaheb et al.,

2012), Gray Level Co-occurrence Matrix (GLCM)

(Öztürk & Akdemir, 2018), Wavelet (Putra, Suciati,

& Wijaya, 2011).

GLCM is a feature extraction method used to

obtain features in the image by calculating the

Angular Second Moment (ASM), contrast, Inverse

Difference Moment (IDM), energy, correlation of co-

occurrence matrix(Mohanaiah, Sathyanarayana, &

GuruKumar, 2013). Based on previous research used

the GLCM method as feature extraction of brain

tumor images for brain tumor classification (Zulpe &

Pawar, 2012) and feature extraction of glaucoma

images for the diagnosis of glaucoma (Karthikeyan

Puspitasari, W., Novitasari, D. and Utami, W.

Batik Classiﬁcation using Texture Analysis and Multiclass Support Vector Machine.

DOI: 10.5220/0008517300650071

In Proceedings of the International Conference on Mathematics and Islam (ICMIs 2018), pages 65-71

ISBN: 978-989-758-407-7

&Rengarajan, 2013). Feature extraction using the

GLCM method has faster calculation and then GLCM

used to recognize the pattern (Mohanaiah,

Sathyanarayana, & Gurukumar, 2013).

Beside GLCM method, there is a Wavelet

Transform method to use features extraction. Chen et

al. (2002) have successfully classified breast tumors

using feature extraction of Wavelet Transformation

and ANN classification. In addition, Rangkuti (2014)

has done research on batik classification using

Discrete Wavelet Transform with duabechies 2 type,

where a type of duabechies 2 is better than other

types. Angelos Tzotos shows that the Support Vector

Machine (SVM) is excellent for object-based image

analysis (Tzotsos & Argialas, 2008). SVM can only

perform binary classification. However, currently,

there is a multiclass SVM approach to solve many

class classification problems. Multiclass SVM with

ECOC approach has been implemented to diagnose

the erythromotropic-squamous disease with high

accuracy results (Prasetyo, 2014). Based on previous

research, batik classification will be performed using

feature extraction with GLCM and feature extraction

results with DWT where classification using

multiclass SVM.

2 LITERATURE REVIEW

2.1 Batik

The batik is a fabric patterned / special motif made by

applying a Malam on the fabric and processed by a

particular process (Musman & Arini, 2011). The

types of batik can be classified based on the method

of manufacture, origin area, and motif (Wulandari,

2012). Batik motif is formed from point, line, and the

plane then becomes an abstract pattern, natural

(natural), geometric and another pattern. Each pattern

of batik has its own philosophical meaning. This

makes batik as a craft that has high artistic value.

Sometimes, some motifs are designed for important

events such as engagements, weddings party,

uniforms, etc.

2.2 Gray Level Co-occurrence Matrix

(GLCM)

Gray Level Co-occurrence Matrix (GLCM) is a

feature extraction method that uses second-order grey

level histogram (Embaugh, 2017). Capture features

based on two parameters, that is distance and angle.

Distance is the pixel difference used for second-order

statistics, an angle formed between pixel pairs. In the

GLCM method, angle orientation is expressed in

degrees. The angular orientation is divided into 4

different angle directions with the 45 ° interval, which

is 0°, 45°, 90°, 135° (Shi & Jeon, 2006). Co-

occurrence direction can be seen in Figure 2. Let 

 represent images of size  and  that have

pixels with L levels and r is the direction vector of

spatial offset. 







defined by the number of

pixels  occurring at offset r to pixel  which can

be expressed as follows





































(1)

where the offset r can be an angle or distance,  

and   . A co-occurrence matrix is

used to get the feature from the image. Harlick's

suggested features include both angle (ASM)

moments, contrast, inverse difference moment

(IDM), energy, correlation(Mohanaiah,

Sathyanarayana, & GuruKumar, 2013).

Figure 2: Co-occurrence Matrix direction for extracting

texture features.

2.2.1 Angular Second Moment (ASM)

ASM is also known as uniformity ASM is related to

energy, where energy is the sum of squares of second

moment GLCM(Suresh, 2012). The highest value is

achieved when the image has excellent homogeneity

when the GLCM elements are all the same

(Mohanaiah, Sathyanarayana, & Gurukumar, 2013).

ASM is calculated using the following formula (2).

Range energy value between [0,1].

 





(2)

where  is matriks co-occurrence,  

and   .

2.2.2 Contrast

Contrast is a measure of the presence of variations in

the pixel grey level of the image. Contrast is

calculated using the following formula (3).

ICMIs 2018 - International Conference on Mathematics and Islam

 

  

  





 



(3)

2.2.3 Inverse Difference Moment (IDM)

IDM is a local homogeneity. IDM has a high value

when the same local gray level and the inverse of the

high GLCM. IDM is obtained from the following

formula (4).

 









 













(4)

2.2.4 Entropy

Entropy is a measure of the randomness of the grey in

the image. Entropy reaches the highest value when

GLCM elements have relatively equal amounts and

have a low cost when the GLCM elements are close

to 0 or 1. Entropy is calculated using the following

formula (5) (Thamaraichelvi & Yamuna, 2016).

 













(5)

2.2.5 Correlation

Correlation is used to calculate the gray linear

dependence of the neighboring pixel. To obtain

correlation values can use the following formula (6).

 

  













 





















(6)

where :







 











(7)







 











(8)





 



  





















(9)







  





















(10)

2.3 Discrete Wavelet Transform

Wavelet Transform method is very influential in the

field of signal analysis, especially in analysis and

image compression. Wavelet is a dangerous method

of image and video compression because of

hasprofessive character in reconstruction. The

wavelet transform is divided into two versions, there

are Continuous Wavelet Transform (CWT) and

Discrete Wavelet Transform (DWT). From the

perspective of computing, the DWT method better

than the CWT method (Embaugh, 2017). DWT

method works multiraisonally and provides

information about the frequency and timing of the

signal. In DWT-2D image processing used wavelet

filter horizontally then vertically to produce four sub-

bands that contain wavelet coefficient value. The type

of wavelet filter is the Low Pass Filter (LPF) and

High Pass Filter (HPF) that evolved from the mother

wavelet. Wavelets develop into several types, there

are haar / db1, duabechies (db2,3,4 .., n), coiflets,

symlets, discrete major, and bioerthogonal)

(Jayaraman, Esakkirajan, & Veerakuma, 2011).

Here's the filtering scheme shown by Figure 3.

Figure 3: Decomposition level 2.

The LL, LH, HL, HH are the result of

decomposition level 1. Lowpasslowpass (LL)

contains the original image approximation

coefficient, lowpasshighpass (LH) contains the

horizontal subband edge coefficient, highpass-

lowpass (HL) contains the vertical subband edge

image coefficient, andhighpasshighpass(HH)

contains the vertical edge coefficient information.

When decomposition of two levels is done, then the

subband that can be decomposed only sub only,

because LL contains information about the image.

From the wavelet decomposition can be

calculated several statistical characteristics by using

the following formula

Batik Classiﬁcation using Texture Analysis and Multiclass Support Vector Machine

 









(11)

  













(12)

 



  

 



 









(13)

 











(14)

2.4 Support Vector Machine

SVM is a binary classification method by dividing

two different classes using the best hyperplane (He,

Wang, Jin, Zheng, & Xue, 2012). But in the real

world problems are often classified into more than

two classes. To resolve these problems can use SVM

multiclass approach. There are two types of data sets

that can be classified using SVM, there are linear data

and non-linear data.

In linear SVM, two different classes are separated

by the best divisor function (hyperplane). Hyperplane

best obtained from the most optimal margin. While

the non-linear data used kernel trick that can map the

training data into a feature vector that has a higher

dimension. There are several kernels that can be used

such as linear kernels, polynomial kernels, and

Gaussian kernels (Shigeo, 2010). Below are some

kernel formulas:

a. Linear Kernel

��,�′=��′,

b. Polynomial Kernel











 







 



c. Gaussian Kernel





































where 







is a pair of two data from all parts of the

training data, parameter  dan  is a constant.

2.4.1 Binary Classification

Two-class classification is done by dividing the data

into two different classes using hyperplane best.

Hyperplane best obtained from the most optimal

margin. The margin is the distance between the

hyperplane with the closest data to the hyperplane of

each class. The data is called a support vector.

2.4.2 Multiclass Classification

Multiclass SVM is used to solve classification

problems of more than two classes. There are three

multiclass approaches, which is one-against-one,

one-against-one and ECOC approaches (Prasetyo,

2014). The one-against-all approach, making binary

classifiers as much as K is then trained to separate the





class vector from the others, for each 



 . Then

each data object is classified into the class where the

greatest decision value is determined (Chih-Wei Hsu,

2002). The second approach, which is one-against-

one with classification binary form as K (K-1) / 2. The

vectors are not the member of the class 



or 



are

ignored when the formation of a binary classifier

(



,



). The last approach, the ECOC approach works

by providing a string of bits called codewords of

length , where  denotes the number of classes.

Then created a binary classification of to predict

every bit codeword. To calculate the predicted result,

look for the closest distance between the codeword

and the classifier by using the Hamming distance

(Prasetyo, 2014).

3 METHODS

Stages of batik classification using the result of

texture analysis with SVM multiclass classification

method shown in Figure 3.

Some stages of batik classification are, datasets,

preprocessing, feature extraction, and classifiers: the

first stage, data collection batik image. The data is

divided into two groups which is training data and

testing data. The information is stored in one file with

the format .jpg.

The second stage, which will be pre-processing

for all data by changing all the image of batik into

gray scales that when extracting features does not get

the effect of RGB / HSV colors. The third stage is

feature extraction. In this research, used two feature

extraction methods, there are Gray Level Co-

occurrence Matrix (GLCM) and Discrete Wavelet

Transform (DWT). For the GLCM method with a

default degree orientation or 0 degrees and a distance

of 1 pixel can be calculated the frequency of gray

pairs appears between pixels in the direction and

distance that has been determined. After kookuren

matrix formed can be calculated statistical

characteristics of the image that is energy, in contrast,

homogeneity, correlation using Equations (2), (3),

(4), (6). While the DWT method using Daubechies 2,

level 3 can be calculated several statistical

characteristics there are energy, entropy, standard

ICMIs 2018 - International Conference on Mathematics and Islam

deviation, and mean using Equation (11), (12), (13)

and (14). The feature will be used for classification

stages. In the classification process used the multi

mile SVM method with the ecoc approach. The stages

in the multiclass SVM classification determines the

characteristic parameter, select the kernel that fits the

data, training data, and the test data to obtain

classification results. In this research using Gaussian

Kernel, The Gaussian Kernel was obtained through

Equation (9). The data will be divided into two, three

and four classes.

Figure 4: Flowchart Batik Classification.

4 RESULTS AND DISCUSSION

In this research, batik will be classified into two

classes which is Parang and Nitik, three classes there

are Parang, Nitik, and Semen, four classes there are

Parang, Nitik, Semen and Buketan. The process of

batik classification uses the GLCM and DWT

methods for feature extraction and the SVM

Multiclass with the ECOC approach to the

classification process.

Before the feature extraction, pre-processing is

required. In pre-processing, color images (RGB) are

converted into grayscale images. This process is used

to simplify the image model so that when extracting

features does not get the effect of RGB / HSV colors.

Image changes to grayscale images shown in Figure

After obtaining a grayscale image, the next step is

the feature extraction process. Feature extraction is

used to get statistical features in the picture. There are

two feature extraction methods used are GLCM and

DWT. In this study using the GLCM method with an

orientation of 0 degrees and a distance of 1 pixel.

(a) (b)

(e) (f)

(g) (h)

Figure 5: (a) Parang batik image, (b) gray scale image of

Parang batik, (c) Nitik batik image, (d) grayscale image of

Nitik batik, (e) Semen batik image, (f) grayscale image of

Semen batik, (g) Buketan batik image, (h) gray scale image

of Buketan.

After determining the direction, then make the co-

occurrence matrix using Equation (1). From the

matrix can be calculated statistical characteristics,

there are energy, contrast, homogeneity and

correlation using Equation (2), (3), (4) and (6). The

samples from feature extraction using GLCM are

shown in Table 1.

While the DWT method using duabechies 2 level

3. From the wavelet decomposition only from the

approximation, coefficients can be calculated

statistical features which is mean, standard deviation,

energy and entropy using Equations (11), (12), (13),

(14). The samples from feature extraction using DWT

are shown in and Table 2.

Batik Classiﬁcation using Texture Analysis and Multiclass Support Vector Machine

Table 1: Sample data feature extraction result using GLCM.

Data

Contrast

Corela-

tion

Energy

Homogeni

2.32750

0.55703

0.07465

0.59638

1.78760

0.73334

0.06928

0.67667

1.83738

0.68679

0.11120

0.65869

2.07442

0.75875

0.08655

0.66997

2.34345

0.59124

0.05418

0.6244

0.28138

0.93332

0.10911

0.86300

1.56753

0.81615

0.11668

0.69613

1.42406

0.82779

0.11793

0.71025

5.17642

0.42063

0.02992

0.50094

0.66446

0.930196

0.08059

0.775365

Table 2: Sample data feature extraction result using DWT.

Data

Energy

Entropy

Mean

Standart

Deviation

0.6538

-1.6E+08

-0.11388

4.200617

2.67966

-7.4E+08

-0.0668

5.263958

1.06521

-2.2E+08

-0.26316

3.304104

0.7446

-1.2E+08

0.04548

5.188262

4.02238

-5.1E+08

0.252135

5.116634

4.22896

-4.5E+08

0.037634

4.344027

0.31044

-1.8E+07

0.003963

1.700911

2.6412

-1.7E+08

-0.12314

3.307191

0.61473

-1.9E+08

-0.0126

3.976344

2.85036

-8.9E+08

-0.26698

7.125455

These features are used as parameters in the

classification. There are two processes in the

classification, the training process, and the testing

process. The ratio of training data and testing data

used were 70%:30%, 75%:25%, and 80%:20%. The

classification process uses the ECOC Multiclass

SVM approach with the Gaussian Kernel. SVM

multiclass classificatory performance is measured

using three measure of performance is the accuracy,

precision, and recall. The result of an experiment with

four different types of data sharing between the two

methods stated in Table 3, Table 4 and Table 5.

Table 3: The results of the two-class classification.

Data

Method

Accuracy

Precision

Recall

70% :

30%

GLCM

92.3 %

91.87%

DWT

73.05%

84.78%

65%

75% :

25%

GLCM

96 %

95%

96.88%

DWT

76 %

86.36%

66.67%

80% :

20%

GLCM

91.3%

88.89%

93.75%

DWT

78.26%

88.1%

64.29%

Table 4: The results of the three-class classification.

Data

Method

Akurasi

Presisi

Recall

70% :

30%

GLCM

86.67%

84.04%

82.92%

DWT

63.33%

46.33%

43.33%

75% :

25%

GLCM

88.89%

90.51%

82.64%

DWT

66.67%

43.81%

43.75%

80% :

20%

GLCM

88.46%

89.58%

82.04%

DWT

67.86%

43.13%

47.62%

Table 5: The results of the four-class classification.

Data

Method

Akurasi

Presisi

Recall

70% :

30%

GLCM

77.14%

82.50%

62.12%

DWT

42.85%

27.78%

30.97%

75% :

25%

GLCM

75.00%

67.19%

DWT

53.13%

39.44%

34.90%

80% :

20%

GLCM

74.19%

77.03%

62.96%

DWT

58.06%

46.18%

38.91%

Based on Table 3, the classification of batik into

two-classes using GLCM based on the distribution of

data sets shows the best results in 75%:25% with 96%

accuracy, precision 95%, and 96.88% recall. From the

experiments show that the GLCM method is perfect

for recognizing each class. While the classification of

batik using DWT showed the best results on the

distribution of data sets 80%: 20% with an accuracy

of 78.26%, precision of 88.1% and recall of 64.29%.

From the experiment shows that the DWT method

can only recognize one class.

Based on Table 4, the classification of batik into

three-classes using GLCM based on the distribution

of data sets shows the best results in 75%: 25% with

an accuracy of 88.89%, precision of 90.51%, and

recall of 82.64%. The experiment shows that the

GLCM method is good for recognizing each class.

While the classification of batik using DWT showed

the best results on the distribution of data sets

80%:20% with an accuracy of 67.86%, precision of

43.13%, and recall 47.62%. The experiment shows

that the DWT method is good enough for recognizing

each class.

Based on Table 5, the classification of batik into

four-classes using GLCM based on the distribution of

data sets shows the best results in 70%: 30% with an

accuracy of 77.14%, precision of 82.50% and recall

of 62.12%. The experiment shows that the GLCM

method is good enough to recognize each class. While

the classification of batik using DWT showed the best

results on the distribution of data sets 80%: 20% with

an accuracy of 58.06%, precision of 46.18% and

recall of 38.91%.

ICMIs 2018 - International Conference on Mathematics and Islam

5 CONCLUSIONS

Based on, the experiment that has been performed,

batik classification into two classes and three classes

get the best results when using GLCM feature

extraction method with an accuracy of 96% for two

type of batik motifs, accuracy of 88.89% for three

kinds of batik motifs and accuracy of 77.14% for four

kinds of batik motifs. This indicates that the feature

extraction using GLCM method is better than the

DWT method to recognize batik pattern based on the

pattern.

ACKNOWLEDGMENTS

A big thank you to Allah SWT for giving blessings

and guidance in the life of the author, UINSA which

has provided an opportunity for the author to seek

knowledge.

REFERENCES

Achjadi, J., 1999. Batik Spirit of Indonesia. Jakarta: PT.

Buku Antar Bangsa.

Babasaheb G. P., Mane, N. N., and Subbaraman, S., 2012.

IRIS Feature Extraction and Classification using

FPGA. International Journal of Electrical and

Computer Engineering (IJECE), Vol 2, No 2, April

2012, 214–222.

Chen, D.-R., Chang, R.-F., Kuo, W.-J., Chen, M.-C., &

Huang, Y. .-L., 2002. Diagnosis of breast tumors with

sonographic texture analysis using wavelet transform

and neural networks. Ultrasound in Medicine &

Biology, 28(10), 1301–1310.

Chih-Wei Hsu, 2002. A comparison of methods for

multiclass support vector machines. IEEE Transactions

on Neural Networks, 12(2), 425–547.

Embaugh, S. E., 2017. Digital Image Processing and

Analysis Applications with MATLAB and CVIPtools

(Third Edition). Boca Raton: CRC Press.

Fanani, A., Yuniarti, A., & Suciati, N., 2014. Geometric

Feature Extraction of Batik Image Using Cardinal

Spline Curve Representation. TELKOMNIKA

(Telecommunication Computing Electronics and

Control), 12(2), 397.

He, X., Wang, Z., Jin, C., Zheng, Y., & Xue, X., 2012. A

simplified multi-class support vector machine with

reduced dual optimization. Pattern Recognition Letters,

33(1), 71–82.

Jayaraman, S., Esakkirajan, S., & Veerakuma, T., 2011.

Digital Image Processing. New Delhi: Tata McGraw

Hill Education Private Limited.

Karthikeyan, S., & Rengarajan, N., 2013. Performance

analysis of gray level cooccurrence matrix texture

features for glaucoma diagnosis. American Journal of

Applied Sciences, 11(2), 248–257.

Moertini, V. S. & Sitohang, B., 2005. Algorithms of

Clustering and Classifying Batik Images Based on

Color, Contrast and Motif. ITB Journal of Engineering

Science, 37(2), 141–160.

https://doi.org/10.5614/itbj.eng.sci.2005.37.2.5

Mohanaiah, P., Sathyanarayana, P., & Gurukumar, L.,

2013. Image Texture Feature Extraction Using GLCM

Approach. International Journal of Scientific &

Research Publication, 3(5), 1–5.

Musman, A., & Arini, A. B., 2011. Batik Warisan

Adhiluhung Nusantara. Yogyakarta: G Media.

Öztürk, Ş., & Akdemir, B., 2018. Application of Feature

Extraction and Classification Methods for

Histopathological Image using GLCM, LBP,

LBGLCM, GLRLM and SFTA. Procedia Computer

Science, 132 (ICCIDS), 40–46.

Prasetyo, E., 2014. Data Mining Mengolah Data Menjadi

Informasi Menggunakan MATLAB. Yogyakarta: Andi.

Putra, R. E., Suciati, N., & Wijaya, A. Y., 2011.

Implementing Content Based Image Retrieval for Batik

Using Rotated Wavelet Transform And Canberra

Distance. Bali International Seminar on Science and

Technology, 1–5.

Rangkuti, A. H., 2014. Content based batik image

classification using wavelet transform and fuzzy neural

network. Journal of Computer Science, 10(4), 604–613.

Shi, Y. Q., & Jeon, B., 2006. Digital Waternarking. Korea:

Springer.

Suresh, A., 2012. Image Texture Classification using Gray

Level Co-Occurrence Matrix Based Statistical

Features. European Journal of Scientific Research,

75(4), 591–597.

Tzotsos, A., and Ardialas, D., 2008. Support vector

machine classification for object-based image analysis.

Final Paper. Retrieved from http://datascienceassn.

org/sites/default/files/A%20Support%20Vector%20M

achine%20Approach%20for%20Object%20Based%20

Image%20Analysis.pdf, accessed on December 8,

2018.

Wulandari, A., 2012. Batik Nusantara - Makna Filosofis,

Cara Pembuatan & Industri Batik. Yogyakarta: Andi.

Zulpe, N., & Pawar, V., 2012. GLCM textural features for

Brain Tumor Classification. International Journal of

Computer Science, 9(3), 354–359.

Batik Classiﬁcation using Texture Analysis and Multiclass Support Vector Machine