RapidOCRLib 0.9.5

RapidOCRSharp

Open source OCR for the security of the digital world

Join our Discord

简体中文 | English

📝 Introduction

RapidOCR is a fully open-source, free, offline-deployable OCR tool that supports multiple platforms and languages. It is designed with a strong focus on extreme speed and broad compatibility.

Supported Languages:
By default, it supports Chinese and English. For additional supported languages, please refer to the documentation:
👉 Model List

Project Background:
Given that PaddleOCR still has room for improvement in engineering and deployment, we aim to simplify and accelerate OCR model inference across various platforms.

We convert PaddleOCR models into highly compatible ONNX format, and provide implementations in multiple programming languages such as Python, C++, Java, and C#, enabling seamless cross-platform deployment and easy integration.

Name Meaning:
"RapidOCR" reflects our vision:

• ⚡ Fast (quick response, easy to use)
• 💡 Efficient (low resource usage, cost-effective)
• 🤖 Intelligent (powered by deep learning for accurate recognition)

We focus on building lightweight yet powerful models, always pursuing speed while ensuring excellent recognition performance.

👉👉 This repository is the C# implementation of the main RapidOCR project:
👉 https://github.com/RapidAI/RapidOCR

If you find this project helpful, please consider giving us a ⭐ Star. Your support means a lot!

👉 Quick Start

The following example demonstrates how to get started using a .NET Console Application.

1. Create a console project

dotnet new console -n ocrdemo
cd ocrdemo

2. Add package reference

dotnet add package RapidOCRLib

3. Add the following code to Program.cs

using Emgu.CV;
using RapidOCRLib;
using System.Diagnostics;

var path = Path.Combine(AppContext.BaseDirectory, "models");

// Initialize model
OcrLite ocrEngin = new OcrLite()
{
    DetPath = Path.Combine(path, "ch_PP-OCRv5_mobile_det.onnx"),
    ClsPath = Path.Combine(path, "ch_ppocr_mobile_v2.0_cls_infer.onnx"),
    RecPath = Path.Combine(path, "ch_PP-OCRv5_rec_mobile_infer.onnx"),
    KeyDicPath = Path.Combine(path, "ppocrv5_dict.txt"),
};

await ocrEngin.InitModels();

// Image to OCR
var demoWillOCRFile = Path.Combine(AppContext.BaseDirectory, "Assets", "demo.png");

// OCR detect
var result = ocrEngin.Detect(demoWillOCRFile, 50);

if (result != null)
{
    Console.WriteLine(result.ToString());

    var img = result.BoxImg.ToBitmap();
    var desFile = Path.Combine(
        AppContext.BaseDirectory!,
        "Assets",
        Path.GetFileNameWithoutExtension(Path.GetRandomFileName()) + ".png"
    )!;

#pragma warning disable CA1416
    img?.Save(desFile!);
#pragma warning restore CA1416

    if (File.Exists(desFile))
    {
        ProcessStartInfo psi = new ProcessStartInfo
        {
            FileName = desFile,
            UseShellExecute = true
        };

        Process.Start(psi);
    }

    Console.WriteLine("Display OCR result:");
    Console.WriteLine(result.StrRes);
}

Console.ReadKey();

⚠️ Note:

• Ensure that model files and images are placed in the correct directories.
• You can also run the RapidOCRConsole project directly, which includes a lightweight PP-OCRv5 model.

⚒️ Project Structure

❓ Parameters & Tuning

OcrLite provides both synchronous and asynchronous methods:

// Sync
public OcrResult Detect(string imgFile, int padding = 50, int maxSideLen = 1024, float boxScoreThresh = 0.5f, float boxThresh = 0.3f,
                     float unClipRatio = 1.6f, bool doAngle = true, bool mostAngle = false);

// Async
public async Task<OcrResult> DetectAsync(string imgFile, int padding = 50, int maxSideLen = 1024, float boxScoreThresh = 0.5f, float boxThresh = 0.3f,
               float unClipRatio = 1.6f, bool doAngle = true, bool mostAngle = false);

Parameter Explanation

If padding is applied, in an automated scenario, the padding must be subtracted to obtain the correct coordinate position.

🎉 Model Download

Download models from: https://www.modelscope.cn/models/RapidAI/RapidOCR/files

Required files:

🎁 Roadmap

• Web demo support
• Table structure recognition
• Automatic model download