Edi Sugiarto, Fikri Budiman, Amiq Fahmi, MY Teguh Sulistyono, Asih Rohmani

Implementation of Discrete Wavelet Transform and Directed Acyclic Graph SVM for Batik Pattern Recognition

Introduction

Implementation of discrete wavelet transform and directed acyclic graph svm for batik pattern recognition. Optimize Indonesian batik pattern recognition using DWT and DAGSVM. This technology improves accuracy by 3% for automatic pattern identification, preserving heritage.

Abstract

Batik as a heritage of the ancestors of the Indonesian nation certainly needs to be preserved so that it continues to be recognized from generation to generation, one of which is by introducing the diversity of its patterns. Efforts to introduce batik patterns can be made, one of which is by implementing technology that can recognize batik patterns automatically based on batik patterns, namely pattern recognition technology. This study aims to optimize batik pattern recognition using the discrete wavelet transform (DWT) and directed acyclic graph SVM (DAGSVM) methods. The stages start from preprocessing, feature extraction, and classification. The study used 310 batik images of 7 different patterns and divided into 240 images for training data and 70 for testing data. DWT method is used in the feature extraction stage while DAG SVM is used in the classification stage. The study was conducted by comparing the accuracy between standard DAG SVM and DAG SVM that has been optimized with DWT and the results of the accuracy test can be proven that adding the DWT method with DAG SVM can increase accuracy by 3%.

Review

This paper addresses a relevant and important problem concerning the preservation and recognition of Indonesian batik patterns, a crucial aspect of cultural heritage. The authors propose an automated pattern recognition system utilizing discrete wavelet transform (DWT) for feature extraction and a directed acyclic graph support vector machine (DAGSVM) for classification. The objective to optimize batik pattern recognition using these established methods is clearly articulated, setting a focused scope for the study. The methodology employed follows a logical sequence of preprocessing, feature extraction, and classification, which is standard for image pattern recognition tasks. The research uses a modest yet reasonable dataset of 310 batik images, comprising seven distinct patterns, with a clear split for training (240 images) and testing (70 images). A key strength lies in the comparative analysis, where the accuracy of the DWT-optimized DAGSVM is directly compared against a standard DAGSVM. The reported result, indicating a 3% increase in accuracy with the integration of DWT, suggests that wavelet-based features are indeed beneficial in enhancing the discriminative power for batik pattern recognition. While the reported 3% improvement is a positive outcome, the abstract lacks the baseline accuracy figure, which would provide greater context for the significance of this gain. It would be beneficial to know the absolute accuracy achieved by both the standard and optimized DAGSVM models. For future iterations, the authors might consider exploring the robustness of their system to common image challenges such as variations in lighting, scale, or slight distortions. Furthermore, benchmarking the proposed DWT-DAGSVM approach against more contemporary deep learning-based methods, which have shown remarkable success in complex pattern recognition, could offer a comprehensive understanding of its performance relative to state-of-the-art techniques.

Full Text

Comments

You need to be logged in to post a comment.