CV Dazzle: Camouflage From Computer Vision by Adam Harvey

Sept. 2012

The Future of the Future

CV Dazzle is an independent project, but it won't be for long.

If you're a coder, computer vision expert, fashion designer, hacker, makeup artist, hair stylist, 3D modeler, privacy enthusiast, activist, fashion designer, or have something to contribute to the project, you should really introduce yourself:

News

Inspiration: Check out these awesome CV Dazzle inspired face paintings from UK artist Celestielle Paint ( http://t.co/3vvRIESC ). Kids can hate big brother too!

Presentation: CV Dazzle goes to Pennsylvania for Philly Tech Week April 20 - 21, 2012

Exhibition: CV Dazzle goes to Switzerland May 5 - Aug 26, 2012. The first design from 2010 will on display at the Voegele Kulturzentrum in Switzerland as part of the DEFENCE exhibition.

Biometrics Challenge: CV Dazzle is going to Rome. I'll be debating the implications of spoofing with leading biometrics researchers on May 10-11. The conference is part of Tabula Rasa, a project funded by the EU to study spoofing attacks on biometrics. May 10 - 11, 2012

Follow me (@adamhrv) for more updates.

Introduction

CV Dazzle™ is camouflage from computer vision (CV). It is a form of expressive interference that combines makeup and hair styling (or other modifications) with face-detection thwarting designs. The name is derived from a type of camouflage used during WWI, called Dazzle, which was used to break apart the gestalt-image of warships, making it hard to discern their directionality, size, and orientation. Likewise, the goal of CV Dazzle is to break apart the gestalt of a face, or object, and make it undetectable to computer vision algorithms, in particular face detection.

Because face detection is the first step in automated facial recognition, CV Dazzle can be used in any environment where automated face recognition systems are in use, such as Google's Picasa, Flickr, or Facebook (see CV Dazzle vs PhotoTagger by Face.com).

Project Overview

This project began as a thesis proposal at the Interactive Telecommunications Program at New York University in the spring of 2010 with the primary objective of thwarting face detection under the guise of high-fashion aesthetics. While there are several obvious approaches to hiding from face detection, some of these can be dismissed. Sunglasses, for example, are a known occlusion which some algorithms account for. And, though functionally effective, wearing masks in public can be illegal. Hoods are popular and effective but make the wearer's intent to hide too obvious. As an alternative, this project explores ways of hiding in plain sight using ambiguously deceptive fashion.

Objective

CV Dazzle opposes the mainstream push towards the widespread adoption of face recognition in order to protect privacy. As the usefulness and popularity of facial recognition grows in commerce and security (currently it's the fastest growing sector of biometrics ), so will the value of privacy. The objective of CV Dazzle is to adapt to our new environment and explore ways of communicating with machines to control our privacy in public.

Design

There is a strong emphasis towards radical-neutrality. The designs used in the first several looks are inspired by both tribal paint and high-fashion aesthetics from the club scene in London. In fact, photos from both were incorporated into the testing algorithms. Surprisingly, many of the more eccentric looks did not fool the face detection algorithms.

To design the looks at left, software was developed that combines interactive drawing and genetic algorithms to detect vulnerabilities in the face detection process. By understanding how face-detection algorithms work, an anti-face can be constructed and used as a guide for creating makeup and hair-styling that foils the face detection process. As a result, your face becomes undetectable to machines yet retains some level of legibility to humans.

Future Work

CV Dazzle is a work in progress. As computer vision matures so will this project. If you would like to get invovled, send me an email. There are many ways to help.

Code (coming soon)

I hope to post the code soon to github. Currently everything is written in Java/Processing using OpenCV1 and is undergoing a migration to OpenCV2.3. Follow me @adamhrv for updates.

Related Projects

The ObscuraCam from the Guardian Project

Tips

While I continue working on a more robust and accessible camouflage, here are a few starter tips from The perilous glamour of life under surveillance.

1. Avoid enhancers: They amplify key facial features.
2. Partially obscure the nose-bridge are: The region where the nose, eyes, and forehead intersect is a key facial feature.
3. Partially obscure the ocular region: The position and darkness of eyes is a key facial feature.
4. Remain inconspicuous: For camouflage to function, it must not be perceived as a mask or disguise.

And a few more tips from an interview with Wired.co.uk

Collaborations

DIS Magazine - How to Hide From Machines

Exhibitions

Videos

Discussion

Interviews

Presentations

Presented at IxDS in Berlin on Feb. 24, 2011
Presented at the 2010 HOPE conference on July 18th

Credits & Thanks

DIS Magazine
Hair: Pia Vivas
Model: Jen
Thesis advisor: Despina Papadapolous
Students and faculty at ITP

Image Use

If you use images from this site or would like to support further development of this project, please consider making a small donation to continued research in developing accessible camouflage from computer vision. Thanks, Adam.
Donate to CV Dazzle

Awards

CV Dazzle is the recipient of a Core77 Design Award

Privacy is normal. Surveillance is suspicious.

Camouflage from Computer Vision