Back to Projects
Web-Based Javanese Script Classifier

Preserving Cultural Heritage Through AI-Powered Handwriting Recognition

8/30/20241 views
Javanese Script Classifier Overview

Problem

Traditional legacies such as the Javanese script are becoming neglected in modern education. As digital communication evolves, fewer people can read or write this ancient script — posing a risk to its long-term survival. Even with available digital archives, accessibility and interactive learning tools remain limited.

The core challenges were:

  • Lack of labeled datasets for handwritten Javanese characters.
  • Limited research on transfer learning applied to low-resource linguistic scripts.
  • Need for an accessible and engaging web interface for public learning.

Solution

The project implemented a TensorFlow-based image classification model using transfer learning to recognize handwritten Javanese characters. This model was deployed as a web application using Streamlit, allowing users to draw characters directly on a canvas and get instant classification results.

Key Features:

  • Interactive Canvas: Users can handwrite characters using a digital pen or mouse.
  • Real-Time Prediction: The model predicts the most likely character based on learned features.
  • Loss and Accuracy Tracking: Implemented multi-epoch visualization for optimization monitoring.
  • Educational Interface: Displays predicted Javanese characters with visual and textual reference.

tldr

  • Situation: The Javanese script was at risk of being forgotten due to limited exposure and lack of digital engagement.
  • Task: Build a tool that modernizes script learning using AI-driven recognition.
  • Action: Developed a TensorFlow classifier trained on custom handwriting data and embedded it into a Streamlit web interface.
  • Result: Created an educational platform combining cultural preservation and AI interactivity.

Technical Highlights

  • Model: TensorFlow deep learning classifier (transfer learning applied).
  • Languages: Python.
  • Frameworks: TensorFlow, Streamlit.
  • Dataset: Custom dataset of handwritten Javanese characters, augmented and normalized.
  • Visualization: Epoch-wise loss curves and inference comparison charts.
  • Deployment: Deployed as an interactive web app for public accessibility.
Javanese Character Training Samples
Training Loss Curves per Epoch
Web Interface with Interactive Drawing

Reflection

This project demonstrated how machine learning can aid cultural preservation beyond mainstream use cases. By applying transfer learning and deploying an accessible web platform, the work bridges the gap between AI research and local heritage education.

Future improvements include:

  • Expanding dataset coverage for compound Javanese characters.
  • Integrating a teaching mode to guide learners through proper stroke order.
  • Deploying as a progressive web app (PWA) for offline use in schools and cultural centers.