A Picture is Worth a Thousand Words, Faces, and Barcodes—The Shape Detection API

Warning: We’re currently working the specification for this API as part of the capabilities project. We’ll keep this post updated as this new API moves from design to implementation.

What is the Shape Detection API?

With APIs like navigator.mediaDevices.getUserMedia and the new Chrome for Android photo picker, it has become fairly easy to capture images or live video data from device cameras, or to upload local images. So far, this dynamic image data—as well as static images on a page—has been opaque, even though images may actually contain a lot of interesting features such as faces, barcodes, and text.

In the past, if developers wanted to extract such features on the client side, for example to build a QR code reader, they had to rely on external JavaScript libraries. This could be expensive from a performance point of view and increase the overall page weight. On the other side, operating systems including Android, iOS, and macOS, but also hardware chips found in camera modules, typically already have performant and highly optimized feature detectors such as the Android FaceDetector or the iOS generic feature detector CIDetector.

The Shape Detection API opens up these native implementations and exposes them through a set of JavaScript interfaces. Currently, the supported features are face detection through the FaceDetector interface, barcode detection through the BarcodeDetector interface, and text detection (Optical Character Recognition, [OCR]) through the TextDetector interface.

Note: Text detection, despite being an interesting field, is not considered stable enough across either computing platforms or character sets to be standardized at the moment, which is why text detection has been moved to a separate informative specification.

Read explainer

Suggested use cases for the Shape Detection API

As outlined above, the Shape Detection API currently supports the detection of faces, barcodes, and text. The following bullet list contains examples of use cases for all three features.

• Face detection

  • Online social networking or photo sharing sites commonly let their users annotate people in images. By highlighting the boundaries of detected faces, this task can be facilitated.
  • Content sites can dynamically crop images based on potentially detected faces rather than rely on other heuristics, or highlight detected faces with Ken Burns panning and zooming effects in story-like formats.
  • Multimedia messaging sites can allow their users to overlay funny objects like sunglasses or mustaches on detected face landmarks.

• Barcode detection

  • Web applications that read QR codes can unlock interesting use cases like online payments or web navigation, or use barcodes for establishing social connections on messenger applications.
  • Shopping apps can allow their users to scan EAN or UPC barcodes of items in a physical store to compare prices online.
  • Airports can expose web kiosks where passengers can scan their boarding passes’ Aztec codes to show personalized information related to their flights.

• Text detection

  • Online social networking sites can improve the accessibility of user-generated image content by adding detected texts as img[alt] attribute values when no other descriptions are provided.
  • Content sites can use text detection to avoid placing headings on top of hero images with contained text.
  • Web applications can use text detection to translate texts, for example, to translate restaurant menus.

Current status

How to use the Shape Detection API

The interfaces of all three detectors, the FaceDetector, the BarcodeDetector, and the TextDetector, are very similar. They all provide a single asynchronous method detect that takes an ImageBitmapSource as an input (that is, either a CanvasImageSource, a Blob, or ImageData).

In the case of FaceDetector and BarcodeDetector, optional parameters can be passed to the detector’s constructor that allow for providing hints to the underlying native detectors.

Note: If your ImageBitmapSource has an effective script origin which is not the same as the document’s effective script origin, then attempts to call detect will fail with a new DOMException whose name is SecurityError. If your image origin supports CORS, you can use the crossorigin attribute to request CORS access.

Working with the FaceDetector

const faceDetector = new FaceDetector({
  // (Optional) Hint to try and limit the amount of detected faces
  // on the scene to this maximum number.
  maxDetectedFaces: 5,
  // (Optional) Hint to try and prioritize speed over accuracy
  // by, e.g., operating on a reduced scale or looking for large features.
  fastMode: false
});
try {
  const faces = await faceDetector.detect(image);
  faces.forEach(face => console.log(face));
} catch (e) {
  console.error('Face detection failed:', e);
}

Working with the BarcodeDetector

const barcodeDetector = new BarcodeDetector({
  // (Optional) A series of barcode formats to search for.
  // Not all formats may be supported on all platforms
  formats: [
    'aztec',
    'code_128',
    'code_39',
    'code_93',
    'codabar',
    'data_matrix',
    'ean_13',
    'ean_8',
    'itf',
    'pdf417',
    'qr_code',
    'upc_a',
    'upc_e'
  ]
});
try {
  const barcodes = await barcodeDetector.detect(image);
  barcodes.forEach(barcode => console.log(barcode));
} catch (e) {
  console.error('Barcode detection failed:', e);
}

Working with the TextDetector

const textDetector = new TextDetector();
try {
  const texts = await textDetector.detect(image);
  texts.forEach(text => console.log(text));
} catch (e) {
  console.error('Text detection failed:', e);
}

Feature detection

Purely checking for the existence of the constructors to feature detect the Shape Detection API doesn’t suffice, as Chrome on Linux and Chrome OS currently still expose the detectors, but they are known to not work (bug). As a temporary measure, we instead recommend doing feature detection like this:

const supported = await (async () => 'FaceDetector' in window &&
    await new FaceDetector().detect(document.createElement('canvas'))
    .then(_ => true)
    .catch(e => e.name === 'NotSupportedError' ? false : true))();

Best practices

All detectors work asynchronously, that is, they are not blocking the main thread 🎉, so don’t rely on realtime detection, but rather allow for some time for the detector to do its work.

If you are a fan of Web Workers (and who isn’t?), the good news is that the detectors are exposed there as well. The detection results are serializable and can thus be passed back from the worker to the main app via postMessage. The demo shows this in action.

Not all platform implementations support all features, so be sure to check the support situation carefully and see the API more like a progressive enhancement. For example, some platforms might support face detection per se, but not face landmark detection (eyes, nose, mouth,…); or the existence and the location of text may be recognized, but not the actual text contents.

Note: This API is an optimization and not something guaranteed to be available from the platform for every user. Developers are expected to combine this with their own image recognition code and take advantage of the native optimization when it is available.

Feedback

We need your help to ensure that the Shape Detection API works in a way that meets your needs and that we’re not missing any key scenarios.

We’re also interested to hear how you plan to use the Shape Detection API:

• Have an idea for a use case or an idea where you’d use it?
• Do you plan to use this?
• Like it, and want to show your support?

Share your thoughts on the Shape Detection API WICG Discourse discussion.

Helpful Links

• Public explainer
• API Demo | API Demo source
• Tracking bug
• ChromeStatus.com entry
• Blink Component: Blink>ImageCapture