Leveraging SIFT Features for Superior SVM-Based Recognition of Hanacaraka Javanese Characters Bambang Yuwono, Mangaras Yanu Florestiyanto, Dessyanto Boedi Prasetyo, Rama Tri Agung
UPN Veteran Yogyakarta
Abstract
Abstract. The Hanacaraka Javanese script has garnered little public interest, and the number of users is declining, partly due to the relatively high difficulty level of teaching and learning Javanese. To help preserve this cultural script, a handwritten character recognition system for the Javanese script has been proposed. Previous research has explored various approaches for the system^s application, such as CNN, KNN, and SVM, each with limitations. This study applies the SVM and feature extraction SIFT method to enhance classification performance. The study uses 2940 records with 20 classes, which have been augmented by seven variations and some preprocessing, precisely resizing and grayscaling. The parameters were optimised by testing 20 combinations of image size, K value, C value, and Gamma value. The best parameters obtained from the test were an image size of 192x192 pixels, a K value of 750, a C value of 3, and a Gamma value of 0.1023. The SVM-SIFT model classification using these parameters achieved high accuracy, with 92.11% on training data and 94.55% on test data. This indicates that the SIFT feature extraction significantly improves the accuracy of the SVM model.
