Improved Magic Square Transform Combined with Double Random Phase Encoding for Iris Recognition

To address the issue of iris template leakage using irreversible transformation, this study proposes a new double iris template protection method that combines an improved magic square transformation with Double Random Phase Encoding (DRPE). Firstly, the user's iris features using the log-Gabor...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2025 4th Asia Conference on Algorithms, Computing and Machine Learning (CACML) s. 1 - 6
Hlavní autoři: Gao, Yanna, Zou, Xue
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 28.03.2025
Témata:
Accuracy
DRPE
Encoding
Encryption
Feature extraction
Fourier transforms
improved magic square
Iris recognition
iris template protection
logistic mapping
Machine learning algorithms
Protection
Vectors
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:To address the issue of iris template leakage using irreversible transformation, this study proposes a new double iris template protection method that combines an improved magic square transformation with Double Random Phase Encoding (DRPE). Firstly, the user's iris features using the log-Gabor transform were extracted, then scrambling iterations and transformation matrix parameters. The iris feature matrix is divided into non-overlapping blocks and scrambled for the first layer of encryption protection using the improved magic square method. Next, in the fractional Fourier transform domain, DRPE is applied to construct an encrypted feature vector. This involves using a user-defined mask M1 and a second mask M2 derived from the user's other iris image, for a second layer of encryption on the scrambled matrix. Finally, encrypted templates are matched for recognition using Hamming distance. Images from the CASIA-IrisV3-Interval, MMU-V1, and CASIA-IrisV4-Lamp iris databases were used to test the algorithm's performance. Results demonstrate that the proposed method achieves an accuracy of 98.93%, can generate a total of 10 60 possible templates, and requires 314 52×624 ×10 60 attempts to recover the original data. This approach meets the standards for irreversibility, revocability, and unlinkability in biometric template protection.
DOI:10.1109/CACML64929.2025.11010947