Feature Extraction using Histogram of Oriented Gradients and Moments with Random Forest Classification for Batik Pattern Detection

Abstract

The preservation of traditional batik patterns, often transmitted orally and through direct practice across generations, faces significant challenges in the modern era. Globalization introduces the risk of cultural homogenization, potentially diminishing the uniqueness and diversity of these patterns. Furthermore, the manual recognition of batik motifs is labor- intensive, time-consuming, and requires specialized expertise, rendering it unsuitable for large-scale preservation initiatives. Consequently, the development of technology-based solutions capable of documenting, analyzing, and recognizing batik patterns with efficiency and precision is imperative for safeguarding this cultural heritage. This study aims to address these challenges by developing an automated system for recognizing batik patterns, focusing on Javanese batik motifs—Kawung, Megamendung, and Parang—which serve as foundational designs for the evolution of batik in other regions. The proposed methodology integrates two feature extraction techniques, Histogram of Oriented Gradients (HOG) and Texture Moments, with the Random Forest machine learning algorithm. The research process encompasses four key stages: pre-processing, feature extraction, classification, and system evaluation, where the accuracy of individual and combined feature extraction methods is analyzed. Experimental results reveal that the HOG method achieves an accuracy of 78.99%, while the Texture Moments method yields 81.88%. Notably, the combination of these two methods enhances system performance, achieving the highest accuracy of 86.23%, representing a 4.65% improvement over the single methods. These findings underscore the efficacy of integrating HOG and Texture Moments with the Random Forest algorithm for automated batik pattern recognition.