Classification of East Sumba Weaving Motifs Based on Texture Using GLCM and SVM Methods

Abstract

East Sumba woven fabric represents a cultural heritage of significant aesthetic and philosophical value, expressed through its diverse array of motifs. However, manual identification of these motifs is prone to high subjectivity, inconsistency, and time-consuming processes. This study aims to develop an automated classification system for East Sumba woven fabric motifs based on texture characteristics, utilizing a combination of the Gray Level Co-occurrence Matrix (GLCM) method and the Support Vector Machine (SVM) algorithm. The dataset comprises 500 digital images of woven fabrics covering five motif variations Baby Crocodile, Chicken, Bird, Dragonfly, and Horse with 100 samples for each. The research workflow encompasses image preprocessing (grayscale conversion), GLCM texture feature extraction (using contrast, correlation, energy, and homogeneity parameters), and classification modeling via an RBF kernel-based SVM algorithm. The dataset was split into 400 training images and 100 testing images. System performance was evaluated using a confusion matrix. The model achieved excellent overall performance, yielding an accuracy of 85.00%, macro-precision of 91.43%, macro-recall of 85.00%, and a macro-F1 score of 82.55%. A slight drop in accuracy was attributed to classification ambiguity in the Chicken motif class, which shares highly similar texture feature values ​​with the Bird motif. Nevertheless, the study demonstrates that the integration of GLCM and SVM methods is highly effective and suitable for implementation to support the automation, digitization, and preservation of East Sumba's cultural heritage.