New: June 2020
Anti Face

This face is unrecognizable to the Viola-Jones Haar Cascade face detection algorithm. (It does not apply to DCNN face detectors)

Computer Vision Dazzle Camouflage

CV Dazzle explores how fashion can be used as camouflage from face-detection technology, the first step in automated face recognition. It is a concept and strategy, not a pattern or product, and it is always designed relative to a specific algorithm and unique to each face.

CV Dazzle, as a concept, can be applied to any computer vision algorithm. There is no one single CV Dazzle design, but many designs for different people for different algorithms. It is encouraged that you create an entirely new look. The looks featured here are for reference only. It is not advised to recreate them.

The name CV Dazzle was inspired by a type of World War I naval camouflage called Dazzle, which used cubist-inspired designs to break apart the visual continuity of a battleship and conceal its orientation and size. Likewise, CV Dazzle uses avant-garde hairstyling and makeup designs to break apart the continuity of a face. Since facial-recognition algorithms rely on the identification and spatial relationship of key facial features, like symmetry and tonal contours, one can block detection by creating an “anti-face”.

NB: I have received many messages regarding the relevance of CV Dazzle in 2020. "Does it work?" is a common question, but needs to be reformatted. CV Dazzle is neither a product nor a pattern. CV Dazzle is a concept and a strategy: you can appear recognizable to people but unrecognizable to machines, existing in a dual state of perception. Most examples here were designed between 2010 — 2013, before neural networks were widely used, and were appropriately designed for the Viola-Jones Haar Cascade method of face detection, but were not designed to be used in 2020 against DCNNs. Therefore you should not expect a design from 2010 to be applicable to an algorithm from 2020. A more helpful question to ask would be "does design X work against algorithm Y?"

Look Book

CV Dazzle Look Book 2010 - present

responsive
Look N° 5 (a)
  • For New York Times Op-Art
  • Model: Bre Bitz
  • Hair: Pia Vivas
  • Makeup: Giana DeYoung
  • Assistant Creative Direction: Tiam Taheri
responsive
Look N° 5 (b)
  • For New York Times Op-Art
  • Model: Bre Bitz
  • Hair: Pia Vivas
  • Makeup: Giana DeYoung
  • Assistant Creative Direction: Tiam Taheri
responsive
Look N° 5 (c)
  • For New York Times Op-Art
  • Model: Bre Bitz
  • Hair: Pia Vivas
  • Makeup: Giana DeYoung
  • Assistant Creative Direction: Tiam Taheri
responsive
Look N° 4
  • For DIS Magazine (2010)
  • Creative direction by Lauren Boyle and Marco Roso
  • Model: Jude
  • Hair: Pia Vivas
responsive
Look N° 3
  • For DIS Magazine (2010)
  • Creative direction by Lauren Boyle and Marco Roso
  • Model: Jude
  • Hair: Pia Vivas
responsive
Look N° 2
  • For DIS Magazine (2010)
  • Creative direction by Lauren Boyle and Marco Roso
  • Model: Irina
  • Hair: Pia Vivas
responsive
Look N° 1
  • For NYU ITP Thesis Presentation (2010)
  • Hair: Pia Vivas
  • Model: Jen Jaffe
  •  
responsive
Look + 1
  • Photo ©Cha Hyun Seok
  • From Coreana Museum of Art Workshop
  • Creative Direction: Kim Hyuna
  • Makeup: Moon Yoo Jin
  • Makeup (Jewelry Details): Bai Yingai
  • Makeup (Stud Details): Jung Ji Moon
  • Hair: Choi Ji Soo, Gu Ye Na
  • Face Chart Design: Kim Ka Hyun, Kim Ye Bin
  • Model: G-Squre Model Academy
  • More info
responsive
Look + 2
  • Photo ©Cha Hyun Seok
  • From Coreana Museum of Art Workshop
  • Creative Direction: Kim Hyuna
  • Makeup: Moon Yoo Jin
  • Makeup (Jewelry Details): Bai Yingai
  • Makeup (Stud Details): Jung Ji Moon
  • Hair: Choi Ji Soo, Gu Ye Na
  • Face Chart Design: Kim Ka Hyun, Kim Ye Bin
  • Model: G-Squre Model Academy
  • More info
responsive
Look + 3
  • Photo ©Cha Hyun Seok
  • From Coreana Museum of Art Workshop
  • Creative Direction: Kim Hyuna
  • Makeup: Moon Yoo Jin
  • Makeup (Jewelry Details): Bai Yingai
  • Makeup (Stud Details): Jung Ji Moon
  • Hair: Choi Ji Soo, Gu Ye Na
  • Face Chart Design: Kim Ka Hyun, Kim Ye Bin
  • Model: G-Squre Model Academy
  • More info
responsive
Look + 4
  • Photo ©Cha Hyun Seok
  • From Coreana Museum of Art Workshop
  • Creative Direction: Kim Hyuna
  • Makeup: Moon Yoo Jin
  • Makeup (Jewelry Details): Bai Yingai
  • Makeup (Stud Details): Jung Ji Moon
  • Hair: Choi Ji Soo, Gu Ye Na
  • Face Chart Design: Kim Ka Hyun, Kim Ye Bin
  • Model: G-Squre Model Academy
  • More info

Advice for Journalists Writing About CV Dazzle or Similar Technologies

  • The looks feautured on this page are archived. They were designed for the Viola-Jones Haar Cascade algorithm. It is not expected that they would work against deep convoluational neural networks. Nor does this website make any claim that they do.
  • The first CV Dazzle designs were made to block face detection in order to prevent further analysis by face recognition. The looks on this page do not make any claim to "trick facial recognition algorithms". They were designed for the vulnerabiliites in the Viola-Jones Haar Cascade "face detection" algorithm. Face detection and face recognition are two completely different algorithms that should not be conflated even though they are often used in the same system. The looks on this page were designed to block face detection, thereby blocking subsequent face recognition algorithms.
  • In order to test the looks on this page with the Viola-Jones haarcascade algorithm, I provide sample code on my personal website at https://ahprojects.com/cvdazzle and encourage you to try it to help understand the algorithm.
  • To test a design against a face detection algorithm, you of course need to first specify which algorithm you are using. Once you know the algorithm, then you can test a look or design relative to that algorithm. Without prior konowledge of the algorithm, it is not really possible to design or test a look or even to say whether CV Dazzle works or not. I encourage you to research the face recognition system you are interested in and gain knowledge of its algorithms before making claims about whether CV Dazzle works not.
  • Journalists and artists should check with a technologist to run an algorithm-specific check before using their design in real world situations
  • And remember, CV Dazzle designs are relative to the algorithm they are being used against. There are hundred of face detection algorithms and hundreds more face recognition algorithms. There are also facial landmark algorithms. Please make sure you specifiy which algorithm it is being designed for.
FAQ
  • Does CV Dazzle still work? Yes. But you need to know more about the algorithm you are targeting. Writing that CV Dazzle "works" or "doesn't" work should always include reference in writing to the algorithm that is being targeted. For most of the looks on this page, the algorithm is the Viola-Jones Haar Cascade face detection algorithm. If you're testing a look, be sure to research and mention the algorithm that you are testing against.
  • Does CV Dazzle block face recognition? If CV Dazzle blocks face detection, then it would also block automated face recogntion. These are completely separate algorithms and should not be conflated. Face recognition requires the bounding box (x,y) coordinates of a face. If these coordinates can not be provided automated face recognition will fail. Again, you may need to research the algorithm you are using. Holding a look up to an iPhone or Android and checking the output does not answer this question.
  • Does CV Dazzle work on dark skin? Yes, probably better as many facial recognition systems were trained with biased datsets that do not include enough data to learn separable visual characteristics of underrepresented faces. But as this site emphases, it is imortant that you know which algorithm you are targeting in order to provide a helpful answer.
  • Does CV Dazzle block face detection? See above. You may need to write code and run a test to check whether your design blocks a specific algorithm. Remember CV Dazzle designs are relative to an algorithm. This cannot be stressed enough. Again, holding a look up to an iPhone or Android and checking the output does not answer this question.
Additional Notes:
  • It is not recommended that you recreate the looks here. These looks are provided for reference, not as a template. Further, these images have likely been added to face detection datasets and neural networks have learned how to detect faces with makeup or facial accessories. It is also not recommended that you post your looks on Instagram or Facebook, as those too will be used to make neural networks stronger.

Style Tips for Reclaiming Privacy

This advice was posted in 2010 and may not be as relevant for deep convoluational neural networks used today.

1
Makeup

Avoid enhancers: They amplify key facial features. This makes your face easier to detect. Instead apply makeup that contrasts with your skin tone in unusual tones and directions: light colors on dark skin, dark colors on light skin.

2
Nose Bridge

Partially obscure the nose-bridge area: The region where the nose, eyes, and forehead intersect is a key facial feature. This is especially effective against OpenCV's face detection algorithm.

3
Eyes

Partially obscure one of the ocular regions: The position and darkness of eyes is a key facial feature.


4
Masks

Avoid wearing masks as they are illegal in some cities. Instead of concealing your face, modify the contrast, tonal gradients, and spatial relationship of dark and light areas using hair, makeup, and/or unique fashion accessories.

5
Head

Research from Ranran Feng and Balakrishnan Prabhakaran at University of Texas, shows that obscuring the elliptical shape of a head can also improve your ability to block face detection. Link: Facilitating fashion camouflage art

6
Asymmetry

Facial-recognition algorithms expect symmetry between the left and right sides of the face. By developing an asymmetrical look, you may decrease your probability of being detected.

Collaborations

From DIS Magazine's How to Hide from Machines

Press

Selected Press
Presentations
  • Upcoming in 2014: Future Everything. Manchester. March 2014.
  • Quintessenz. Wien, Austria. 2013.
  • SMART Design. 2013
  • Tabula Rasa: Spoofing, Anti-Spoofing Workshop. Rome. 2012.
  • IxDS. Berlin. 2011
  • New York Times Design Series. 2010.
  • Philly Tech Week. 2012
  • HOPE Hackers Conference. 2010

Contact

CV Dazzle is developed by Adam Harvey, an artist whose work explores the impacts of surveillance technologies. CV Dazzle was originally developed a masters thesis project while at New York University in 2010.

OpenCV Visualized

OpenCV Face Detection

OpenCV is one of the most widely used face detectors. This algorithm performs best for frontal face imagery and excels at computational speed. It's ideal for real-time face detection and is used widely in mobile phone apps, web apps, robotics, and for scientific research.

OpenCV is based on the the Viola-Jones algorithm. This video shows the process used by the Viola Jones algorithm, a cascading set of features that scans across an image at increasing sizes. By understanding how the algorithm detects a face, the process of designing an "anti-face" becomes more intuitive.

Click image for video: http://vimeo.com/12774628

Image Use

All images ©2010-2020 Adam Harvey unless otherwise noted. Please contact prior to use.