platform_ocr

A high-performance, local OCR package for Dart and Flutter that leverages native OS APIs. This package directly utilizes the built-in OCR capabilities of Windows and Darwin (macOS/iOS) for optimal speed and accuracy without external dependencies like Tesseract.

Features

Blazing Fast: Uses native OS-optimized OCR models.
Local Processing: No data is sent to the cloud, ensuring user privacy and offline support.
Platform Native: Utilizes Windows.Media.Ocr on Windows and Vision.framework on Darwin.
Native Assets: Seamless integration using Dart Native Assets for automated building and linking.

Platform Support

Platform	Native API	Status
Windows	`Windows.Media.Ocr` (WinRT)	✅ Implemented
iOS	`Vision.framework`	✅ Implemented
macOS	`Vision.framework`	✅ Implemented

Note

Android is not supported as this package focuses on OCR features built directly into the OS.

Getting Started

Prerequisites

Windows: Visual Studio 2022 with "Desktop development with C++" and "C++/WinRT" components.
macOS/iOS: Xcode with latest SDKs.

Installation

Add platform_ocr to your pubspec.yaml:

dependencies:
  platform_ocr: ^1.0.0

Since this package uses Native Assets, ensure you are using a compatible Flutter version (3.13+) and have the necessary build tools installed.

Usage

import 'package:platform_ocr/platform_ocr.dart';

void main() async {
  final ocr = PlatformOcr();

  // From a file
  final textFromFile = await ocr.recognizeText(OcrSource.file(File('image.png')));
  print('Recognized: $textFromFile');

  // From memory
  final textFromMemory = await ocr.recognizeText(OcrSource.memory(uint8List));
  print('Recognized: $textFromMemory');
}

Internal Architecture

Windows

Implemented using C++/WinRT with a C ABI wrapper (src/windows/ocr_cabi.cpp). Images are decoded in Dart using package:image to ensure correct RGBA8 format delivery to the native engine.

Darwin (iOS/macOS)

Implemented using direct Objective-C interop with VNRecognizeTextRequest from the Vision framework. It supports both path-based and memory-based (NSData) image processing.

Roadmap

Support for detection level (Fast vs. Accurate).
Language specific hints.
Bounding box information for recognized text.
Transition to Windows Imaging Component (WIC) for native image decoding on Windows.

License

BSD 3-Clause License