The Artificial Neural Networks (ANN) for Batik Detection Based on Textural Features

Abstract

The problem in this research is the difficulty in distinguishing batik motifs from non-batik motifs for people who do not understand batik motifs. The classification of batik motifs can be done by processing digital images of batik motifs using artificial neural network methods for the classification of batik image motifs. This study aims to utilize artificial neural networks to distinguish batik motifs and non-batik fabric motifs. Several important steps are needed, namely the process of acquiring batik and non-batik images, pre-transforming batik and non-batik images to gray scale forms, texture feature extraction in gray scale images and detection of motifs using networks artificial nerve. Image acquisition is done by collecting batik and not batik images from several different motifs. The preprocessing process is done by changing the image size to 256 x 256. Sixteen texture features are used, namely Angular Second Moment (ASM), contrast, correlation and Inverse Different Moment (IDM) with 45 ° angle intervals, namely 0 °, 45 °, 90 ° and 135 °. Processing data sets is divided into 70 percent as training data and 30 percent as testing data. Artificial neural network models used in this research use the Backpropagation learning algorithm by comparing the Scaled conjugate gradient algorithm (trainscg) training method and the Levenberg-Marquardt algorithm (trainlm) training method. The results obtained for the accuracy of the batik detection model using the Scaled conjugate gradient algorithm (trainscg) training method were higher with an accuracy value of 84.12%, compared to the Levenberg-Marquardt algorithm (trainlm) method by 86.11%.