Handwriting Classification of Sundanese Script Using LBP Feature Extraction and CNN

The Sundanese script is an important element of West Java's cultural heritage, as a community that upholds and preserves culture and art in Indonesia as an active participation, where what can be done is to strengthen and preserve it through digital means. This study proposes a new classification system for Sundanese script handwriting using the Convolutional Neural Network (CNN), either with or without the integration of Local Binary Pattern (LBP) for feature extraction. The data used in this study is image data of 1,600 images of Sundanese script divided into 40 different classes. The results of this study show the effectiveness of Convolutional Neural Network in recognizing complex patterns in handwritten manuscripts. The model trained only with the Convolutional Neural Network achieved an accuracy of 92.11% and had a loss value of 0.7429, far surpassing the hybrid model that included the Local Binary Pattern (LBP) which only achieved an accuracy of 71.21%. These findings suggest that Convolutional Neural Networks with scenarios without the addition of Local Binary Patterns, offer a more efficient and computationally effective approach to the classification of Sundanese scripts. This study not only supports the potential use of CNNs in script recognition but also contributes to broader efforts in the preservation of digital culture.