Building a web-based image markup system

Turker working on an iPad

A web-based service like Amazon Mechanical Turk needs a web-based interface for turkers to crowdsource data on fish shape. Existing software to digitize images requires a separate download, and most of it runs only on Windows. Distributing this software to hundreds of crowdsourced workers and ensuring it works on their computers can be quite a challenging task.

To that end, we’ve had to develop an image digitization interface using only technologies that work in your browser. Specifically, we use the HTML5 canvas element, which is flexible enough to allow arbitrary graphics to be drawn in your browser window, but also gives us the power to use Javascript to record the marks that turkers then submit to our servers.

The elegance of using the web as a platform to drive our crowdsourcing effort is that as long as you have a way to browse the internet, you should be able to contribute your work to our research. Our interface is agnostic to technology choice: we’ve tested it on an iPad and it works quite well.

Click here to see what the interface looks like, or visit GitHub to peek at the source code.

Up next: I discuss the accuracy of crowdsourced landmarks.