Inspiration

During Facebook Hackathon (3-10 May 2021), according to Judging Criteria: One major point was on concerning privacy. To obey the rules of single submission only, this work has not submitted to the hackathon, We switched to another updated and high-tech project. To officially announce the original implementation please follow us.

What it does

There are multiple data poisioning technique regarding face recognition algorithms in ML, AI and deep learning. Up to date methods and regarding possible filters are as follows:

  1. According to FAWKES Reverse Engineering Technique (https://arxiv.org/pdf/2002.08327.pdf): Version 1.1. Placing glitters, decorations like hearts, stars and so on with face filter on the major 5 or 6 points of the face to prevent face recognition algorithms to detect the exact id. Version 1.2. Placing exactly fawkes just as a filter. (ps. This copyright may belong to FAWKES so a collabrative work is necessary)

  2. According to LOWKEY Reverse Engineering of Technique (https://arxiv.org/pdf/2101.07922.pdf): Version 2.1. Placing distortive gifs like moving/waving/flowing water, rainbow, stars, stardust, rains, fruits or pertubating gifs or eddy current gifs etc on the major 5 or 6 points of the face to prevent face recognition algorithms to detect the exact id. Version 2.2. Placing exactly LOWKEY just as a filter. (ps. This copyright may belong to LOWKEY so a collabrative work is necessary)

  3. According to "Hanxun Huang, Xingjun Ma, Sarah Monazam Erfani, James Bailey and Yisen Wang" Research Paper (https://openreview.net/pdf?id=iAmZUo0DxC0): Version 3.1. Placing exactly "Huang et.al"s data poisining methodology as a filter. (ps. This copyright may belong to "Hanxun Huang, Xingjun Ma, Sarah Monazam Erfani, James Bailey and Yisen Wang" so a collabrative work is necessary)

What's next for Face Protective Filters

As more data according to eyes are collected during pandemic, Solely eye based possible candidates can be gathered as well by minority high tec and big data owners. For future, if this would be spreaded: Version 4 would be eye related distruptive techniques are planned that changes the shape of the eye or adding cartoon eye alone just to protect users face data and allow them to share their pictures publicly and freely.

©Copyright Nazenin Gure© Patent Pending Invention date: 04.05.2021 Devpost public date: 09.May.2021 at 00:00 (GMT+3) e-mail proof date regarding the idea: "8 May 2021 15:02" (recorded)

References:

Research paper: https://arxiv.org/pdf/2002.08327.pdf

https://sandlab.cs.uchicago.edu/fawkes/

Their online page that allows to try online:

https://dev.scrb.ai/

Here the secod version, another research team developed further on fawkes and this one is called “lowkey”.

https://arxiv.org/pdf/2101.07922.pdf

This is the online try out version of lowkey:

https://lowkey.umiacs.umd.edu/

This is the ultimate research that leads ai to skip the image :

https://openreview.net/pdf?id=iAmZUo0DxC0

Source:

https://www.technologyreview.com/2021/05/05/1024613/stop-ai-recognizing-your-face-selfies-machine-learning-facial-recognition-clearview/

Built With

  • sparkar
Share this project:

Updates