Last year just before the Chrome Dev Summit, Miguel Casas came up to me and
showed me something that blew my mind: Face Detection in the
browser using the Shape Detection
API. Shortly after
that Barcode Detection was added that allowed me to
update my QR Code scanner so that I no longer
had to include a massive (albeit awesome) port of a QR scanning library.
The Shape Detection API is still in development, and neither the FaceDetection
nor the Barcode Detection API’s are available outside experimentations (you
need to enable “Experimental Web Platform features” in chrome://flags) but it is
a very exciting space to watch and see another platform capability being opened
up to developers and users on the web.
The latest addition is the Text Detection
API that will
take an image and scan it for readable text. The video at the top of this
article is a great example (note, I stole some of the code from Miguel but put
my own spin on it, notably the synthesis part.)
The model is exactly the same as the Face and Barcode Detection APIs, you get an
image (either an img, canvas or ImgData object) and pass it into a instance of
the type of detector you want to use. You can the process the results and
perform some action on the data (for example draw on the image where the item
was detect). In this case the results are an Array of
DetectedText
which you can use to extract what text was detected.
This currently works only on Chrome Canary for Android, but if you want to
experiment check out the code
and the demo and the process is not too
painful although my code is incredibly hacky. I quite like this demo, it detects
text in the image, draws a box around the text and then when the user clicks
inside the bounding box, it will read the text back to the user.
The API is not amazingly complex. If you wanted to implement something yourself
you can follow these steps:
1. Get access to the camera
Query the list of mediaDevices and select the first camera that is front-facing
(note: there are better ways to do this).
2. Capture a full resolution frame
3. Create a TextDetector and start detection
4. Process the results
For each item that is detected and element will appear in the array that is
passed to the Promise returned from the detect function. You can then itterate
over this array, find where they are positioned in the image, and get access to
the data detected.
Yup, I am excited!
This API opens up so many interesting possibilities for users such as easier and
broader access to assistive technologies for parsing content in images;
Real-time translation of text in images; extracting urls for slides at
conferences (totally going to try and get a web app ready for Google IO that
does this), are just a few examples that quickly spring to mind.
What would you do with this API?