CNN-Based Image Processing for Recognizing Balinese Characters on Palm Leaves

Balinese script is an integral part of Balinese cultural heritage, traditionally inscribed on palm leaves. The preservation of this ancient script faces challenges due to technological advancements and changing cultural dynamics. This paper presents a CNN-based image processing approach for recognizing Balinese characters on palm leaves. The proposed method employs Convolutional Neural Networks (CNN) to identify and classify Balinese script from digitized lontar manuscripts. By applying techniques such as RealESRGAN for image resolution enhancement, Otsu thresholding for optimal segmentation, and morphological operations for image refinement, we have improved the recognition accuracy of Balinese characters. Our experimental results, using models like ResNet50, MobileNetV2, and DenseNet169, VGG16 and InceptionResNetV2 demonstrate significant stability in character recognition, with MobileNetV2 achieving an accuracy of $99.73 \%$. This study advances the accessibility and preservation of Balinese script, ensuring that this cultural heritage continues to be recognized and appreciated by future generations.