Batik Image Classification Using Treeval and Treefit as Decision Tree Function in Optimizing Content Based Batik Image Retrieval

April 14, 2019 | Author: Heri Budi | Category: Wavelet, Statistical Classification, Information Retrieval, Computer Vision, Shape
Share Embed Donate

Short Description



 Available  Availabl e online at www.sciencedi

ScienceDirect  Procedia Computer Science 59 (2015 ( 2015)) 577 – 583

International Conference on Computer Science and Computational Intelligence (ICCSCI 2015)

Batik image classificat classification ion using treeval and treefit as decision decision tree function in optimizing content based batik image retrieval 1


Abdul Haris Rangkuti , Zulfany Erlisa Rasjid , DJunaidi Santoso



School of Computer Science, Bina Nusantara University, Jakarta, Indonesia  Email : [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected]  [email protected] 


This research is to increase the percentage of similarity and to increase the speed of the r etrieval of characteristic characteristic of batik image which is the texture and shape. In order to obtain an optimal result, the classification process is performed using a decision tree with treeval and treefit function, where the value used is the result of the image feature extraction. For this image image extraction, the values that originate from the approximation approximation coefficient that uses the wavelet transform method deubecheuss level 2 and i nvariant movement. The research is performed on 7 types of pattern and 225 images. The result using using 5 types of batik patterns namely lereng, parang, kawung, nitik and truntum using using 20 test data on each pattern, pattern, has a similarity percentage above 80 - 85 percent. For 2 other patterns which which is mega mendung and ceplok ceplok using 10 data on each pattern, has a similarity percentage percentage above 30 – 40 percent only. Based Based on the result, further research research is required to using other other methods and functions. © 2015 Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license © 2015 The Authors. Authors. Published Published by Elsevier Elsevier B.V. (  ). Peer-review under responsibility of organizing committee of the International Conference on Computer Science and Computational Intelligence Peer-review under responsibility of organizing committee of the International Conference on Computer Science and Computational (ICCSCI 2015). Intelligence (ICCSCI 2015)  Keywords : Batik, treeval, treefit, approximation approximation coefficient, Wavelet Transform, Transform, Invariant Invariant moment , moment ,

1. Backgrou Background nd

In order to support the development of Indonesia’s culture especially in preserving batik cloth, it is necessary to perform researches that is related to the batik pattern characteristic. Batik is a cultural heritage not because of the batik itself but because of the art in making the batik. The occurrence of the problem in claiming the batik culture is partly caused by the lack of arwareness of our nation on the importance of the batik culture preservation. To prevent this problem from happening, it is required to have a complete documentation documentation on Indonesia’s Indonesia’s batik. In existing research on Batik Batik image retrieval, currently it is  based on colour and shape characteristic and only a few research is using shapae and texture characteristic. In fact, all batik  patterns have symbolic meaning that contains information of the batik image, ima ge, especially on its shape and texture char acteristic. For that reason this research is using using an approach approach involving two extraction extraction features of Batik Image Retrieval (BIR). 19-21, Content Based Batik Image Retrieval Retrieval is an approach for image retrieval based based on the information contained in the Batik image itself such as colour, shape shape and the texture of the image. CBIR comprises of the following steps, preprocess, preprocess, pattern extraction, 24,17 indexing and image retrieval . An image retrieval system is a system to retrieve information in a form of an image by measuring the similarity precentage of  the image query that is input by the user and the image stored in the database 13,15,16. The problem in content based searching searching

1877-0509 © 2015 Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (  ). Peer-review Peer-revie w under responsibility of organizing committee of the International Conference on Computer Science and Computational Intelligence (ICCSCI 2015) doi:10.1016/j.procs.2015.07.551 doi: 10.1016/j.procs.2015.07.551


Abdul Haris Rangkuti et al. / Procedia Computer Science 59 (2015) 577 – 583

system is to find a feature that can represent unique characteristic of the image, so that it can be used accurately to identify images. The visual feature that can be extracted from the image data are texture, colour and shape [16]. In relation to the batik image, the texture feature is an important feature because of the ornaments on the batik cloth can be seen as a different texture composition. This research however, is also focused on the shape and colour feature of the batik image [18][20..21]. The main focus of this research on batik image retrieval system is to obtain an optimal result in obtaining a pattern similarity which is more relevant to the intented image. Therefore this research will be developed not only to find the texture feature but also the shape similarity through the concept of CBIR on batik image27,28. Some research has been performed by developing a retrieval system, based on contents of Batik image, some using the 25 Generalized Hogh Transform method to identify a specific pattern in a batik image . In earlier research it is also developed a retrieval system concept based on Kodebook which is developed using keyblock structure to encode and decode a batik image [28]. An image retrieval concept based on the batik image content that has a specoal characteristic on the image using filter logGabor and colour histogram has also been developed. 1.1 Scope of the Research The scope of this research is : A. The research object is focused on 7 batik pattern which is ceplok, lereng, parang, nitik, truntum, kawung, and mega mendung. This research is focused on the texture and shape characterisic. In the retriecal process, classification process is  performed first using the treefit function to generate the decision tree and the treeval function to form the classification. The research object is batik image which is in a form of an image database and image query with the following specification: having a jpg format and 200 x 200 pixel in size, and the image that is input or used as the image query can be an image with no specific sized. 1.2 Objectives and Benefits The objective of this research is: 1. To develop an accurate image classification method in order to be able to increase the capability in performing batik image processing. 2. To increase the accuracy and speed in the batik image processing by performing classification process using tree fit and treeval functions. 3. To facilitate the batik image retrieval processing from image query that has been grouped into classes based on pattern characteristic in the form of a decision tree using treeval function. The benefits of this research is as follows: A. By developing a method for image extraction and classification, the accuracy of the image retrieval based on characteristic will be increased. B. The result of this research is expected to obtain a high accuracy and high performance of the batik recognition process.


2.1 Treefit and Treval Decision Tree Treefit (x,y) function is to create a decision tree to predict the response y which is predicted by the x value. x can be a marix with a value for prediction. y is also a vector used as a response of n or an array of character that contain class name n and it is  based on value in the x column. Treeval function is used to create the classification or regression from the decision tree produced by the treefit function, including a matrix X on the prediction value. In order to classify and perform regression based on the creator value (yfit), which is based on the response from one point to obain the prediction valu, and the classification treebased on the class number where tree is to relate the point with the ith data (X(i)) and convert a number of points into class name 4. 2.2 Invariant Moment To identify an object in an image, segmentation process frequently has problems with regards to the object’s position, object rotation and changes in the objects’ scale. Changes or rotation in position, different sizes of objects, whether it is small or large causes error in identifying that particular object. Moment is able to represent an object in many ways such as area, position, orientation and other defined parameters. The purpose using this method is to obtain invariant moments from all objects from a Batik image. Every Batik image will have 7 values of invariant moments. All 7 (seven) value of invariant moments will be used 10,20,21 for identification process .

2.3 Texture and Shape Characteristic Calculation

 Abdul Haris Rangkuti et al. / Procedia Computer Science 59 (2015) 577 – 583


At this step, texture feature extraction using type debecheus 2 wavelet transform method in order to obtain decomposition on each Batik image making use of subband LL. In order to obtain an optimal feature extraction, several texture characteristic is such as mean, standard deviation, entropy, correlation, contrast, energy and skewness is used on each Batik image. The representation of the image characteristic is as follows:  Mean : Shows the dispersion of an image including the determination of gray intensity.  Energy : Image characteristic resulted from the image decomposition is obtained by calculating the energy contained in each subband.  Entrophy : Basically, entropy is performed to measure the diversity and the intensity of the image. Entropy resulted from wavelet can be regarded as the characteristic of an image. Entropy shows the variety of the measurement of shapes. Large entropy value for images with uniform degree of gray tr ansition and small value if the image structure varies.  Standard Deviation : Shows the grey intensity distribution. All characteristics will be calculated for each Batik image, resulting in 4 (four) characteristic value related to each Batik image. It is expected that with these 4 characteristic values, extraction process can be more accurate and optimal including to facilitate 1,2,20,21,29 the classification process that will be used in the next step . 2.4

Collection and Selection of Batik Image There are 2 rules in the selection of Batik image; the image condition and the image pattern. For the Batik Image selected to  be analysed, it is important to have a clear shape and texture, whereas for image pattern to be analysed, there are 12 types of Batik image patterns. The collection of the Batik images is performed by several ways such as Batik image acquisition from the internet, direct from digital photo and from magazines related to Batik images. The Batik images collected will be converted into  jpg extension, at the same time performing resizing and mormalization. This process is performed in order to obtain the dimension which is easy to view visually and accurately processed by the system, either 200 x 200 or 300 x 300. 3. Research Framework

Collecting And Selecion of Image batik


Feature Extraction

Feataure Extraction using texture

Collectiong and Selection of Batik Image


Measurement Similarity

Image Classification Using Treefit and Treeval Fuction (Decision Tree)

Preprocessing using graylevel dan Canny Detection

Class of Baitk Image Feataure Extraction using Shape

Measurement Of Image Similarity

Input of batik Image

Result of Batik Image Similarity

Figure 1. Research Framework : Classification Process using Treeva and treefit as function Decision Tree 3.1 Pre processing 3.1.1 Grayscale Process Grayscale process is the process to convert colored images to become a grayscaled image. Basically a color image consists of 3 layers which is R- layer, G-layer and B-layer. The grayscale process will convert 3 layers R, G dan B to become 1 layer and the result will be a grayscaled image. In this image there are no colors, having only the degree of grayness. To convert the color image that has r, g and b to grayscale color with a value s. The conversion can be performed by averaging r, g and b. Therefore there will be no more colored image, all are grayscale images.


Abdul Haris Rangkuti et al. / Procedia Computer Science 59 (2015) 577 – 583

3.1.2 Edge detection algorithm using Canny Edge detection is a process to determine the point l ocation which forms the edges of the objects. Edge detection on an image is a process that form edges from image objects. This method can detect the edges or lines which form image objects and clarification on the edges. Sobel, Prewitt, Robert, Laplacian of a Gaussian, Canny, and other methods are used to detect the 21,25 edges of the images. Canny algorithm has become a standard algorithm t o detect edges and used in many researches . 3.2 Feature Ekstraction The result of existing experiments, starting from grayscale processing, binary and canny processes and the result of invariant moment’s calculation is shown in figure 2.

Figure 2 Preprocessing of Parang pattern and Feature Ekstration using Invariant Moment

3.3 Wavelet Transform (Texture Feature Extraction) Wavelet method is a mathematical function that converts the original image to become an image in the frequency domain, where further can be divided into different subband frequency component. Each component is studied using resolution which is according to scale. According to Sydney (1998), Wavelet is a small wave that has the capability to group image energy, concentrated on a group of small coefficient, whereas the other coefficients only contains a small amount of energy which is trivial and can be ignored without reducing the Information value. The implementation of concept of wavelet transformation on 1,7,8,10,20,14 the image decomposition is adapted using wavelet character related to the object in study . With several researches the feature extraction on Batik image using wavelet transformation supports the Batik Image Retrieval based on Image features, with 20,21,14 sub image frequency researched focusing on low low subband . In general, the process starts from preprocessing until extraction using the wavelet transform method and followed by calculating the feature parameter on each of the Batik pattern. The process is shown in Figure 3.

Prerocessing with grayscale process

Classification using Treeval and Treefit as decision tree fuction

BinerisasI Process

Edge detection

Feature extraction using wavelet transform

To calculate texture characteristic which It is the result process from decomposition process in LL koeficient such as energy, mean, entrophy and standard of d eviation.

Figure 3. Chart showing the process of Batik image fr om preprocessing until Feature extraction. This chart shows the process start from preprocessing to feature extraction process. Preprocessing step started with grayscale  process and binary processing. Then followed by side detection process using the canny algorithm. The next step is feature extraction process using the deubecheus 2 level 2 t ype wavelet transform method. Wavelet will perform the image decomposition 13,15  by obtaining the subband LL value . The result of the decomposition is the approximation coefficient, which will be followed

 Abdul Haris Rangkuti et al. / Procedia Computer Science 59 (2015) 577 – 583


 by calculating 4 features in the feature extr action process. The result of this calculation will become the input for the classification process using treeval dan treefit. 3.4 Decision Tree using Treeval dan Treefit unction After obtaining the coefficient resulted fro  the image extraction process using the wavelet transfor m, classification process 4 is performed using treeval and treefit f unction. The processes are as follows : Load coefficient image 1, image 2 ... image t = treefit(kls, image n);% Create decision tr  e sfit = treeval(t,kls); % Find class tasks sfit = t.classname(sfit); % Obtain class nefficientame kls(strcmp(sfit, imagen)); % Image Coeffici nt placement on each class

The result of experimental from the processing a number of motive batik image 100 50 0

Figure 4 : The result of Exp riment using some batik pattern and the output of the ex eriment

Comparisonofthemethods andalgo rithmsofpreviousstudies

70 60 50 40 30 20 10 0 No Classification

Simplle Classification using FNN

the numberof batik motif st d ie d

Classification using Treeval and treefit

T he av er age o f A ccur acy p er ce nt age in C BI R p ro ce ss

Figure 5 : Comparison of using some method and algorithm to optimal the CBI

Abdul Haris Rangkuti et al. / Procedia Computer Science 59 (2015) 577 – 583


In the figure 4.0 describes the results of experiments on 7 of motive batik image. From Experiment results illustrated, the  percentage accuracy for image retrieval base on the example image. Each pattern has 10 to 12 types of sample images that have almost the same pattern. In the figure 5.0 describes the results of previous research by the author, the number of batik motif studied including the results of the percentage accuracy for the CBIR process. 5. Conclusion Based on the experiments performed, the following can be concluded: 1. The result obtained from the prototype application, the highest precision reached is 90% meaning that the identification of Batik image is very high. On some image pattern which is more difficult to be identified such as mega mendung and ceplok, the precision is between 30-40 %. This will be a concern for the researcher to find other methods or concepts that can increase the similarity percentage. 2. By implementing the treeval and treefit as decision tree function, the process is 0.2-0.3 seconds faster, where previously the  performance is between 0.7 – 0.8 seconds using the same computer configuration. 3. In the previous study, some retrieval process is performed through classification and some are not. However, classification using Fuzzy Neural Network (FNN) and applying the treeval and treefit functions proved that the result is better. 4. This research will be further concducted in order to identify some images that has a more complex characteristics or shapes such as recognizing batik pattern, blood cells, and tumor cells. References 1.

Ajay KS, Tiwari S, VP Shukla ,2012, ”Wavelet Base Multi Kelas Image Classification using Neural Network”, International journal of Computer


Balamurugan and Anandhakumar P, May 2009, “Neuro-Fuzzy Based Clustering Approch For Content Based Image Retrieval Using 2D-Wavelet


Breiman, L., J. Friedman, R. Olshen, and C. Stone. Classification and Regression Trees. Boca Raton, FL: CRC Press, 1984.


Chein C dan H.Ren, An Experiment-Based Quantitative and Comparative Analysis of Target Detection and Image Classification Algorithms f or


Gonzaga, A., de Franca, C. A., & Frere, A. F. (1999, March). Wood texture classification by fuzzy neural networks. In Electronic Imaging'99 (pp. 134-143).

Application”, vol 37 – No.4 Transforn”, International Journal of Recent Trend in Engineering vol.1 No.1

Hyperspectral Imagery, IEEE Trans. on Geoscience and Remote Sensing, 38(2), 2000. International Society for Optics and Photonics. 6.

Hazra D, “Texture Recognition with combined GLCM,


Jawale S,“Gabor wavelet transform co occurent matrix based texture feature for content based image retrieval, International Journal of Engineering

Wavelet and Rotated Wavelet Features”, International Journal of Computer and Electrical

Engineering, Vol.3, No.1, February, 2011 Research and Applications (IJERA)”, National Conference on Emerging Trends in Engineering & Technology,2010 8.

Joseph B, P. Darwin, 2012, “Multi Wavelet for Image Retrival Based On Using Texture and Color Querys”, IOSR Journal of Computer Engineering (IOSRJCE) ISSN: 2278-0661, ISBN: 2278-8727Volume 6, Issue 6 (Nov. - Dec. 2012), PP 10-13


Liantoni, F., Ramadijanti, N., & Rosyid Mubtada'i, N. (2010). Klasifikasi Daun Dengan Centroid Linked Clustering Berdasarkan Fitur Bentuk Tepi Daun. EEPIS Final Project.


Lotfi, M., Solimani, A., Dargazany, A., Afzal, H., and Bandarabadi, M. (2009, March). Combining wavelet transforms and neural network s for image classification. In System Theory, 2009. SSST 2009. 41st Southeastern Symposium on (pp. 44-48). IEEE.


Kokare M, Biswas PK, and Chatterji B N, 2007, "Texture image retrieval using rotated wavelet filters," Pattern Recogn. Lett., vol . 28, pp. 1240-1249,


Mukane, S. M., Bormane, D. S., & Gengaje, S. R. (2011). Wavelet and Co-occurrence Matrix based Rotation Invariant Features for Textu re Image Retrieval using Fuzzy Logic. International Journal of Computer Applications, 24.


Mojsilovic A., Popovic M.V. and D. M. Rackov, “On the selection of an optimal wavelet basis for texture characterization ”, IEEE Transactions on Image Processing, vol. 9, pp. 2043–2050, December 2000.


Neuronales, M. R. (2007). Image Retrieval Based on Wavelet Transform and Neural Network Classification. Computación y Sistemas, 11(2), 143-156.


Park, S. B., Lee, J. W., & Kim, S. K. (2004). Content-based image classification using a neural network. Pattern Recognition Letters, 2 5(3), 287-300.


Reddy PV , Satya Prasad K , 2011, “Multiwavelet Based Texture Features for Content Based Image Retrieval”, International Journal of Computer Science


Pratikaningtyas D dkk, 2010, “ Klasifikasi Batik Mengunakan Metode Transformasi Wavelet ”, Paper Skirpsi, UNDIP.


Rangkuti AH, Bahaweres, R. B., & Harjoko, A. (2012, Dec.). Batik image retrieval based on similarity of shape and texture characteristics. In Advanced

and Technology, JNTU Kakinada, AP, India

Computer Science and Information Systems (ICACSIS), 2012 International Conference on (pp. 267-273). IEEE. 19.

Rangkuti, H. A., Harjoko, A., & Putro, A. E. (2014). Content based batik image retrieval. Journal of Computer Science, 10(6), 925.


Rangkuti, A. H. (2014). Content Based Batik Image Classification Using Wavelet Transform And Fuzzy Neural Network. Journal of Computer Science,10(4), 604-613.


Rallabandi, V. R., & Subramanyam Rallabandi, V. P. (2008). Rotation-invariant texture retrieval using wavelet-based hidden Markov trees.Signal Processing,88(10), 2593-2598.


Shidi, T. A. P., & Suyoto, S. (2011). New Edge Detection Method for Indonesian Batik. Jurnal Buana Informatika, 2(1).


Smeulders, A. W. M. , dkk. (2000). Content-based Image Retrieval at The End of The Early Years. IEEE PAMI, 22(12), 1349-1380


Sengur, A. (2009). Color Texture Classification Using Wavelet Transform and Neural Network Ensembles. Arabian Journal for Science & Engineering (Springer Science & Business Media BV), 34.

 Abdul Haris Rangkuti et al. / Procedia Computer Science 59 (2015) 577 – 583



Sanabila HR and Manurung R, 2009, “Recognition of Batik Motifs using the Generalized Hough Transform”, University of Indonesia.


Osadebey M A, Integrated Content-Based Image Retrieval using texture, shape and spatial information, Umea University, Umea Sweden, Feb, 2006


Sengur, A. (2009). Color Texture Classification Using Wavelet Transform And Neural Network Ensembles. Arabian Journal for Science & Engineering (Springer Science & Business Media BV), 34.


Wahyudi, Azurat A, Manurung M, and Murni A, 2009, “Batik Image Reconstruction Based On Codebook and Keyblock Framework”, University of


Yildizer, E., Balci, A. M., Jarada, T. N., & Alhajj, R. (2012). Integrating wavelets with clustering and indexing for effective content-based image retrieval.

Indonesia. Knowledge-Based Systems, 31, 55-66.

View more...


Copyright ©2017 KUPDF Inc.