Efficient Iris Recognition via Polar Representation and Radial Stripe Attention

Trong-Thua Huynh^1,*, De-Thu Huynh², Cong-Sang Duong¹, Hong-Son Nguyen¹, Quoc H. Nguyen³, Lam-Thanh Tu⁴
1 Faculty of Information Technology II, Posts and Telecommunications Institute of Technology, 11 Nguyen Dinh Chieu Street, Sai Gon Ward, Ho Chi Minh City, Viet Nam
2 School of Computer Science & Engineering, The Saigon International University, 16 Tong Huu Dinh Street, An Khanh Ward, Ho Chi Minh City, Viet Nam
3 Institute of Digital Technology, Thu Dau Mot University, 06 Tran Van On Street, Phu Loi Ward, Ho Chi Minh City, Viet Nam
4 Advanced Intelligent Technology Research Group, Faculty of Electrical and Electronics Engineering, Ton Duc Thang University, 19 Nguyen Huu Tho Street, Tan Hung Ward, Ho Chi Minh City, Viet Nam
* Corresponding Author: Trong-Thua Huynh. Email: email

Computer Modeling in Engineering & Sciences https://doi.org/10.32604/cmes.2026.080616

Received 13 February 2026; Accepted 27 April 2026; Published online 18 May 2026

Download PDF

Abstract

Deep iris recognition models are often trained on Cartesian grids, whereas iris texture follows a concentric structure with angular periodicity. This representational mismatch can weaken rotation robustness and limit pupil-to-limbus context modeling, while many pipelines still rely on accurate segmentation masks. We propose RadialFormer, an efficient mask-free iris recognition framework that performs representation learning directly in the polar domain. The pipeline first estimates pupil/iris parameters (cx,cy,rin,rout) using a percentile radial-gradient operator with anatomical constraints, and then applies a crop-based polar transform to obtain a compact 64×512 unwrapped iris map. To better match polar geometry, we introduce Learnable Polar Position Encoding (LPPE) with separable radial–angular embeddings, where Fourier terms in the angular branch enforce continuity at θ=0/2π. We further propose Radial Stripe Window Attention (RSWA), which computes self-attention within full-height radial stripes and uses modular angular shifting to preserve circular consistency. Trained end-to-end with batch-hard triplet loss under P×K sampling, RadialFormer achieves 99.04% TPR@1%FPR with 0.48% EER on CASIA-V4-Lamp, and 93.63% TPR@1%FPR with 2.92% EER on CASIA-V4-Interval. Ablation and cross-dataset evaluations further validate the contributions of polar processing, LPPE, and RSWA and demonstrate robust generalization across acquisition conditions. Under the same input resolution, RadialFormer reduces computation by about 3.5× compared with a standard transformer baseline while maintaining competitive recognition accuracy.