tesseract-wasm

A WebAssembly build of the Tesseract OCR engine for use in the browser and Node.

tesseract-wasm can detect and recognize text in document images. It supports multiple languages via different trained models.

👉 Try the demo (Currently supports English)

Features

This Tesseract build has been optimized for use in the browser by:

• Stripping functionality which is not needed in a browser environment (eg. code to parse various image formats) to reduce download size and improve startup performance. The library and English training data require a ~2.1MB download (with Brotli compression).

• Using WebAssembly SIMD when available (Chrome >= 91, Firefox >= 90, Safari ??) to improve text recognition performance.

• Providing a high-level API that can be used to run web pages without blocking interaction and a low-level API that provides more control over execution.

Setup

1. Add the tesseract-wasm library to your project:

   npm install tesseract-wasm

2. Serve the tesseract-core.wasm, tesseract-core-fallback.wasm and tesseract-worker.js files from node_modules/tesseract-wasm/dist alongside your JavaScript bundle.

3. Get the training data file(s) for the languages you want to support from the tessdata_fast repo and serve it from a URL that your JavaScript can load. The eng.traineddata file supports English for example, and also works with many documents in other languages that use the same script.

Usage

tesseract-wasm provides two APIs: a high-level asynchronous API (OCRClient) and a lower-level synchronous API (OCREngine). The high-level API is the most convenient way to run OCR on an image in a web page. It handles running the OCR engine inside a Web Worker to avoid blocking page interaction. The low-level API is useful if more control is needed over where/how the code runs and has lower latency per API call.

Using OCRClient in a web page

import { OCRClient } from 'tesseract-wasm';

async function runOCR() {
  // Fetch document image and decode it into an ImageBitmap.
  const imageResponse = await fetch('./test-image.jpg');
  const imageBlob = await imageResponse.blob();
  const image = await createImageBitmap(image);

  // Initialize the OCR engine. This will start a Web Worker to do the
  // work in the background.
  const ocr = new OCRClient();

  try {
    // Load the appropriate OCR training data for the image(s) we want to
    // process.
    await ocr.loadModel('eng.traineddata');

    await ocr.loadImage(someImage);

    // Perform text recognition and return text in reading order.
    const text = await ocr.getText();

    console.log('OCR text: ', text);
  } finally {
    // Once all OCR-ing has been done, shut down the Web Worker and free up
    // resources.
    ocr.destroy();
  }
}

runOCR();

Examples and documentation

See the examples/ directory for projects that show usage of the library in the browser and Node.

See the API documentation for detailed usage information.

See the Tesseract User Manual for information on how Tesseract works, as well as advice on improving recognition.

Development

Prerequisites

To build this library locally, you will need:

• A C++ build toolchain (eg. via the build-essential package on Ubuntu or Xcode on macOS)
• CMake
• Ninja

The Emscripten toolchain used to compile C++ to WebAssembly is downloaded as part of the build process.

To install CMake and Ninja:

On macOS:

brew install cmake ninja

On Ubuntu

sudo apt-get install cmake ninja-build

Building the library

git clone https://github.com/robertknight/tesseract-wasm
cd tesseract-wasm

# Build WebAssembly binaries and JS library in dist/ folder
make lib

# Run tests
make test

To test your local build of the library with the example projects, or your own projects, you can use yalc.

# In this project
yalc publish

# In the project where you want to use your local build of tesseract-wasm
yalc link tesseract-wasm