Descriptive Camera shuns photos for text images
Prototype, created for a class on computational cameras, makes you use your imagination by printing out text descriptions of what it sees, rather than the regular photos you'd expect from a cam.
A picture may be worth a thousand words, but Matt Richardson has figured out how to make 20 words worth a picture. Richardson is the creator of the Descriptive Camera, a camera that prints out text descriptions of what it sees, rather than actual photos.
There's a human element to making this work. Point the camera (in this case, a USB Webcam) at a scene. Take a picture. The picture is sent to Amazon'soutsourcing service. A human writes up a quick description and sends it back. The camera prints it out using a tiny thermal printer.
Here's an example. The camera took a photo of an old building in bright light. The result is a printout reading, "This is a faded picture of a dilapidated building. It seems to be run down and in need of repairs."
Richardson describes the inspiration for the camera in an e-mail to Crave: "It started as a 'what if there were a camera that could record what was happening in the photo as searchable text?' I made this camera to explore that 'what if?'"
Richardson estimates that the camera cost about $200 in parts, but developing the code to make it work took a lot of time. The price for each use of Mechanical Turk is $1.25. That's roughly equivalent to the cost of a Polaroid photo.
The Descriptive Camera was created for a class on computational cameras that Richardson took through New York University this spring.
The prototype is designed to play with the idea of metadata. Our cameras often attach lots of information to photos, such as time, date, location, camera settings, and type of camera. Now we finally have a camera that gives us the most important metadata information of all: what it just took a picture of.