Catch up on the latest AI articles

New Face Identity Transformer With Password-based Facial Anonymization And Recovery, Privacy And Accessibility

New Face Identity Transformer With Password-based Facial Anonymization And Recovery, Privacy And Accessibility

Face Recognition

3 main points 
✔️ Proposed Face ID Transformer for anonymization and recovery with passwords
✔️ Anonymization and decryption in a single network, taking into account the memory requirements of the camera 
✔️ 
Introduced a 1:1 password/face ratio, so even a wrong password will generate a face image, making it impossible for hackers to notice the anonymization

Password-conditioned Anonymization and Deanonymization with Face Identity Transformers
written by 
Xiuye Gu, Weixin Luo, Michael S. Ryoo, Yong Jae Lee
(Submitted on 26 Nov 2019 (v1), last revised 30 Sep 2020 (this version, v4))
Comments: Accepted at ECCV2020
Subjects: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Image and Video Processing (eess.IV)
 
  

overview

Cameras, such as smartphones and security cameras, are becoming more and more popular at a rapid pace. However, there is a growing concern about the leakage of personal information (especially facial information) contained in those images.

There have been reports of techniques to edit and anonymize facial images, but many of these techniques are masking techniques, such as blurring the original image. However, this makes it impossible to see the original image. Some anonymization techniques have been reported that leave traces of the original image, such as techniques that retain only the behavioral information of the original image, but again, the face remains anonymous. These can certainly keep the images safe, but you can't see who is actually doing what, because the images are edited.

For example, if an incident occurs and you want to check the security camera afterward, there is no clue and the camera is useless. Also, even if the privacy of a video image taken by an individual is protected when it is saved, it is meaningless if it is not restored when the image is reviewed. Therefore, anonymization technology must be able to restore the original image if necessary, while preserving the privacy of the video image.

In this paper, we address this problem by using a new Face Identity Transformer We propose the following. An overview is shown below.

First, a password is embedded in the original input images (Inputs) to create an anonymized image (Anonymization). Then, if the correct password is entered, the original image is restored, and if the wrong password is entered, a different image is generated (deanonymization).

Therefore, you can protect your privacy in a way that allows the original image to be recovered, and furthermore, you can build in a mechanism that prevents hackers from realizing that you've entered the wrong password. In other words, even if the wrong password is entered, a facial image is generated, so the hacker can be fooled into thinking that the hack was successfully completed. The framework also allows for anonymization and deanonymization in a single network, taking into account the memory limitations on the camera.

To read more,

Please register with AI-SCHOLAR.

Sign up for free in 1 minute

OR

If you have any suggestions for improvement of the content of the article,
please contact the AI-SCHOLAR editorial team through the contact form.

Contact Us