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.
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
-
Add the tesseract-wasm library to your project:
npm install tesseract-wasm
-
Serve the
tesseract-core.wasm,tesseract-core-fallback.wasmandtesseract-worker.jsfiles fromnode_modules/tesseract-wasm/distalongside your JavaScript bundle. -
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.traineddatafile 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:
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
74 Dec 27, 2022
15 Aug 21, 2022
20 Aug 26, 2022
6.8k Jan 3, 2023
6.1k Jan 9, 2023
6 Dec 17, 2021
18.9k Dec 30, 2022
13.2k Jan 2, 2023
130 Jul 14, 2022
27.3k Jan 4, 2023
46 Jan 10, 2022
107 Dec 26, 2022
9 Dec 9, 2022
1 Jan 3, 2022
3 Jul 20, 2022
30.1k Jan 3, 2023
11 Sep 20, 2022
18 Dec 10, 2022
5 Dec 1, 2022