Cancelable color face recognition using trinion gyrator transform and randomized nonlinear PCANet

For the analysis of color images, the single-channel processing method fails to consider the internal correlation between color components, while the quaternion matrix representation is not compact enough. To avoid these, this paper introduces trinion gyrator transform for color face image and devel...

Full description

Saved in:
Bibliographic Details
Published in:Multimedia tools and applications 2024, Vol.83 (22), p.61435-61449
Main Authors: Shao, Zhuhong, Wang, Xue, Li, Bicao, Zhang, Yu, Shang, Yuanyuan, Ouyang, Junlin
Format: Article
Language:English
Subjects:
Accuracy
Algorithms
Biometrics
Color imagery
Computer Communication Networks
Computer Science
Correlation
Cryptography
Data encryption
Data Structures and Information Theory
Face recognition
Facial recognition technology
Feature recognition
Fourier transforms
Matrix representation
Mobile commerce
Multimedia
Multimedia Information Systems
Principal components analysis
Quaternions
Special Purpose and Application-Based Systems
Track 3: Biometrics and HCI
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:For the analysis of color images, the single-channel processing method fails to consider the internal correlation between color components, while the quaternion matrix representation is not compact enough. To avoid these, this paper introduces trinion gyrator transform for color face image and develop a cancelable recognition scheme by jointing with randomized nonlinear PCANet. Firstly, color face images are precoded into trinion matrices and followed by a random binary amplitude mask for non-reversible. The nonlinear trinion correlation is proposed for selecting the optimal ratio. Next, the sampled matrix is modulated by using trinion gyrator transform and logistic-based random phase mask for increasing discriminability. Afterwards, the randomized nonlinear principal component analysis network is employed for extracting features. The recognition accuracy of the proposed algorithm on VIS, Aberdeen, GT and YMU datasets are up to 99.00%, 99.23%, 99.68% and 97.88%, which outperforms several existed methods. On top of that, the correlation index of generated face templates is no more than 0.26, indicating the revocability and security of the proposal.
ISSN:1573-7721
1380-7501
1573-7721
DOI:10.1007/s11042-023-17905-2