Aksharize OCR

Aksharize is a lightweight full-stack OCR application designed to recognize Malayalam text from signboards, street signs, and public displays, and provide readable English transliterations.

The project consists of:

1. FastAPI Backend (Python)
   Decodes uploaded images, performs OpenCV preprocessing, runs Tesseract OCR, and transliterates Malayalam text into readable English.

2. Flutter Frontend (Mobile)
   Allows users to capture images using the camera or select images from the gallery, preview them, upload them to the backend, and view OCR and transliteration results.

---

Screenshots

Home Screen

---

Image Selected

---

OCR Result

---

Project Structure

aksh/
├── backend/
│   ├── app/
│   │   ├── main.py
│   │   ├── routes/
│   │   │   └── ocr.py
│   │   └── services/
│   │       ├── ocr_service.py
│   │       └── transliteration_service.py
│   ├── requirements.txt
│   └── venv/
│
└── aksharize_app/
    ├── lib/
    │   ├── main.dart
    │   ├── home_screen.dart
    │   └── api_service.dart
    └── android/

---

Features

• Malayalam OCR using Tesseract
• Malayalam to English transliteration
• Kerala place-name dictionary overrides
• Camera image capture
• Gallery image selection
• FastAPI backend
• Flutter mobile frontend
• JSON API responses
• OpenCV image preprocessing
• Cross-platform support

---

Quick Start

1. Backend Setup (FastAPI)

Prerequisites

Python

Install Python 3.10 or newer.

Verify installation:

python --version

or

python3 --version

---

Install Tesseract OCR

macOS

Install Tesseract and language packs:

brew install tesseract
brew install tesseract-lang

Verify Malayalam support:

tesseract --list-langs

Expected output should include:

eng
mal

---

Ubuntu / Debian Linux

sudo apt update

sudo apt install tesseract-ocr

sudo apt install tesseract-ocr-mal

Verify installation:

tesseract --version

Verify language availability:

tesseract --list-langs

Expected output:

eng
mal

---

Windows

Download Tesseract from the UB Mannheim builds:

https://github.com/UB-Mannheim/tesseract/wiki

During installation:

• Select Malayalam language support.
• Install to the default location:

C:\Program Files\Tesseract-OCR

Add Tesseract to your PATH:

C:\Program Files\Tesseract-OCR

Verify installation from Command Prompt:

tesseract --version

Verify Malayalam language support:

tesseract --list-langs

Expected output:

eng
mal

---

Backend Installation

Navigate to the backend folder:

cd backend

Create Virtual Environment

macOS / Linux

python3 -m venv venv

Activate:

source venv/bin/activate

---

Windows

Create environment:

python -m venv venv

Activate (Command Prompt):

venv\Scripts\activate.bat

Activate (PowerShell):

.\venv\Scripts\Activate.ps1

---

Install Dependencies

pip install -r requirements.txt

---

Run Backend

uvicorn app.main:app --reload --port 8000

Backend URL:

http://127.0.0.1:8000

Swagger Documentation:

http://127.0.0.1:8000/docs

---

2. Frontend Setup (Flutter)

Prerequisites

Install Flutter SDK:

https://flutter.dev/docs/get-started/install

Verify installation:

flutter doctor

Ensure all required dependencies are installed.

---

Install Dependencies

Navigate to the Flutter application:

cd aksharize_app

Fetch packages:

flutter pub get

---

Run Application

Start an emulator or connect a physical device.

Run:

flutter run

---

Backend URL Configuration

For Android Emulator:

http://10.0.2.2:8000

This is already configured in the application.

For a physical device, update the API base URL in:

lib/api_service.dart

Example:

const baseUrl = "http://192.168.1.100:8000";

Replace with the local IP address of the machine running the FastAPI server.

---

OCR & Transliteration Pipeline

OCR Preprocessing

The OCR pipeline:

1. Receive uploaded image.
2. Decode image using OpenCV.
3. Convert image to grayscale.
4. Process using Tesseract OCR.
5. Extract Malayalam text.

Current Tesseract configuration:

Language: mal
PSM: 8
OEM: 3

Optimized for:

• Street signs
• Place names
• Shop boards
• Public signage

---

Transliteration Engine

The transliteration system uses two stages:

Stage 1 — Dictionary Overrides

Known Kerala place names are mapped to their commonly accepted English spellings.

Examples:

Malayalam	English
തൃശൂർ	Thrissur
കൊച്ചി	Kochi

This ensures outputs match real-world signage and common usage.

---

Stage 2 — Fallback Transliteration

If no dictionary match exists, the application falls back to the indic_transliteration library to generate a phonetic English transliteration of the OCR result.

---

Future Improvements

• Support for additional languages
• OCR correction and post-processing improvements

---

License

MIT License