Facial Image Privacy Protection Based on Principal Components of Adversarial Segmented Image Blocks

The features in facial images, which are utilized for a variety of technological applications, pose a significant privacy concern for users. This paper proposes a method for protecting privacy in facial images based on the principal components of adversarial segmented image blocks. Generative advers...

Full description

Saved in:
Bibliographic Details
Published in:IEEE access 2020, Vol.8, p.103385-103394
Main Authors: Yang, Jingjing, Liu, Jiaxing, Wu, Jinzhao
Format: Article
Language:English
Subjects:
adversarial samples
Face recognition
Facial image privacy protection
Feature extraction
generative adversarial network
Generators
Image quality
Image segmentation
Linear algebra
Object recognition
Perturbation
Perturbation methods
Principal component analysis
principal components
Privacy
Security management
Target recognition
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The features in facial images, which are utilized for a variety of technological applications, pose a significant privacy concern for users. This paper proposes a method for protecting privacy in facial images based on the principal components of adversarial segmented image blocks. Generative adversarial network parameters are compressed by segmenting the facial images into blocks and extracting the principal components of the segmented image. The generator and discriminator in the generative adversarial network then generate images similar to the original facial images; the facial images generated by the generator, as-driven by the target recognition network, markedly different from the original facial images. As the generator, discriminator, and target recognition network compete with each other, minor perturbation is added to the principal components of the facial images to protect the users' privacy and prevent distinct face-related features of the images from being easily extracted. Experimental results show that the proposed method outperforms other similar methods in terms of generated image quality, operation speed, and target recognition network accuracy.
ISSN:2169-3536
2169-3536
DOI:10.1109/ACCESS.2020.2999449