Arabic OCR (C++23 + ONNXRuntime + OpenCV)

Arabic | 简体中文

---

🌍 English

Introduction

Arabic OCR is an open-source project for Optical Character Recognition (OCR) of Arabic text, built with C++23, ONNXRuntime, and OpenCV. The goal is to provide a fast, lightweight, and portable OCR engine for Arabic script.

Licensed under Apache-2.0, community contributions are welcome!

Features (Planned)

• Deep learning inference with ONNXRuntime
• OpenCV preprocessing (grayscale, binarization, skew correction, etc.)
• Arabic character recognition (initial focus: printed text)
• Clean C++ API for easy integration
• Command-line tool for image → text

Build & Install

Dependencies

• C++23 compiler (GCC 13 / Clang 16 / MSVC 2022+)
• ONNXRuntime
• OpenCV 4.x
• CMake 3.20+

Build

windows_x64_msvc:

git clone https://github.com/lona-cn/ArabicOCR.git
cd ArabicOCR/ArabicOCR-Infer
.\scripts\build-windows_x64_msvc.bat

C++ example

#include <filesystem>
#include <iostream>

#include <opencv2/highgui/highgui.hpp>
#include <windows.h>

#include "ArabicOCR.h"


int main(int argc, char* argv[])
{
    SetConsoleOutputCP(CP_UTF8);
    std::cout << std::format("cwd:{}", std::filesystem::current_path().string()) << std::endl;
    auto infer = arabic_ocr::InferContext::Create().value();
    auto ocr = arabic_ocr::OCR::Create(*infer, "assets/det/inference.onnx", "assets/rec/inference.onnx").value();
    cv::Mat mat = cv::imread("assets/imgs/arabic-1.png");

    auto results = ocr->BatchOCR({mat});
    for (const auto& result : results)
    {
        for (const auto& text_box : result)
        {
            std::cout << std::format("{} {}", text_box.confidence, text_box.text) << std::endl;
        }
    }
    return 0;
}