Sundanese Script Optical Character Recognition

Table of content

• Introduction
• Feature
• How it works
• Setup and Installation

Introduction

Sundanese script OCR, made using YOLO for object detection, and LSTM for transliteration.

Feature

• Individual character detection using YOLO.
• Transliteration to Latin using Seq2Seq LSTM/BiLSTM, based on YOLO detection result.

Flowchart

1. Image is uploaded into the main webapp for cropping
2. Webapp send image to pre-trained YOLO model for detection.
3. Webapp recieve respond from YOLO containing detection result.
4. Webapp process the detection result before sending it to transliteration model for transliteration.
5. Transliteration model repsonds with the transliteration result.

Installation-and-Setup

1. Clone this repository
2. Install the required dependencies:

pip install -r requirements.txt

3. Train model using this Dataset: Sundanese Script Dataset

4. Deploy your trained model on roboflow

5. Insert your own Roboflow API key in the code On App.py:

roboflow_client = InferenceHTTPClient(
    api_url="https://detect.roboflow.com",
    api_key="API PLACEHOLDER" # add your own API key
)

6. Run the application:

python app.py

7. Open your browser and go to http://127.0.0.1:5000/
8. On another terminal, go to Transliteration_model-bcnr directory
9. Run the transliteration model:

python serve.py