Skip to end of metadata
Go to start of metadata


by C. Chris Erway


The Internet, once thought of as providing deeper knowledge and understanding, instead now provides us with an amazingly powerful tool for pursuing shallowness. As we mold ourselves and our relationships to fit "social networks" (the machine's simplified, database-centric model of our society---a schema describing the personal lives of millions of people, in a few hundred lines of SQL) we find ourselves hammered shallower as a result, as if to better fit the 2D plane upon which web pages are projected.

We spend hours indulging strange, new, voyeuristic and narcissistic urges on Facebook and MySpace, urges that we could hardly imagine having or quenching a decade ago. And yet, the technology exists today to endlessly and effortlessly browse photos and check the "status" of long-forgotten friends, lovers and colleagues not heard from in years. Though we may likely never see many of these "friends" in person ever again, we (consciously or not) perform for them, with each new item that we add to our "activity stream" and feed to the machine. In this age everyone can be a celebrity, at least among their peers; users of social networks play both voyeur and exhibitionist (or, author and reader) online.

Similarly, the web has also provided new ways of following those celebrities among us that everyone knows (though the line between fame ex machina and "real" fame is hard to draw these days, now that "going viral" is a legitimate route to stardom: just ask Soulja Boy or Tila Tequila). Aggressive new paparazzi blogs such as TMZ and Perez Hilton (and sites like Twitter, which provide the illusion of friendship with the stars we "follow") allow us to retrieve up-to-the-minute photos and updates about celebrities, pushing the privacy frontier further each day.

This work asks the question: what would the Internet look like if it were not so shallow; what if the online voyeur and exhibitionist in us were suppressed? What if our Facebook friends could no longer see our most recent profile picture update? What if the photos of our friends and favorite celebrities online were blurred and anonymized beyond recognition? What happens when Facebook becomes Facelessbook?

This experience is provided interactively, by filtering all images through a web proxy as they are transferred to the user's browser, and searching each image for faces (using the open source computer vision library OpenCV). Any areas detected as belonging to a face are blurred, and a new image is sent on in place of the original. All this occurs transparently and automatically, without control by the user. The intended effect is to de-emphasize outsized personalities, thwart voyeuristic tendencies, and ideally foster deeper, more meaningful modes of communication.

In a way, the experience can be imagined as viewing the "social network" from the machine's point of view. In this view, a user's outward appearance is much less important than her data/metadata (and related opportunities for targeted advertising, market research, and profit that lay behind better understanding of the "social graph").

Finally, a note on the technology: proxies like these are commonly used to evade censorship and confer anonymity on a user's browsing activity (e.g., by users in China or Iran, or P2P users). However, here it is not the proxy's user that is being anonymized, but all other people that are anonymized from view.

Usage Instructions

To use Defaced, you must configure your browser to use a HTTP Proxy so that it may route all its traffic through the Defaced proxy server. In Firefox, this is under Preferences: Advanced: Network: Connection Settings: HTTP Proxy. In Safari, go to Preferences: Advanced: Proxies: Change Settings. Enter the following details:

Manual proxy configuration
HTTP Proxy server:
Port: 3333
User: faceless
Password: book

Please remember to turn off the proxy server when you are done. You may also find yourself needing to hold down SHIFT and click Reload to make your browser display fresh images (with or without blurring). (Note: Computer vision is a young field, and while the face detection algorithms in OpenCV have been "trained" on thousands of faces, occasionally some slip through its grasp. These issues remain an area of active research, but the current version of my code seems to have most difficult on very small representations of faces.)

Harper's: The Serfdom of Crowds (excerpt from You are Not a Gadget)

OpenCV computer vision library, used with Python as in this example
Squid caching proxy, used as in this image-filtering example
ImageMagick, open source image library, for blurring
Web pages, Python code



(names truncated for privacy)

Perez Hilton

Note Perez tends to draw on paparrazzi photographs; here those images are further altered.


Talking Points Memo

Brown CS Faculty

Enter labels to add to this page:
Please wait 
Looking for a label? Just start typing.