TY  - EJOU
AU  - Huynh, Trong-Thua 
AU  - Huynh, De-Thu 
AU  - Duong, Cong-Sang 
AU  - Nguyen, Hong-Son 
AU  - Nguyen, Quoc H. 
AU  - Tu, Lam-Thanh 

TI  - Efficient Iris Recognition via Polar Representation and Radial Stripe Attention
T2  - Computer Modeling in Engineering \& Sciences

PY  - 
VL  - 
IS  - 
SN  - 1526-1506

AB  - 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 <i>RadialFormer</i>, an efficient mask-free iris recognition framework that performs representation learning directly in the polar domain. The pipeline first estimates pupil/iris parameters <mml:math id="mml-ieqn-1"><mml:mo stretchy="false">(</mml:mo><mml:msub><mml:mi>c</mml:mi><mml:mi>x</mml:mi></mml:msub><mml:mo>,</mml:mo><mml:msub><mml:mi>c</mml:mi><mml:mi>y</mml:mi></mml:msub><mml:mo>,</mml:mo><mml:msub><mml:mi>r</mml:mi><mml:mrow><mml:mtext>in</mml:mtext></mml:mrow></mml:msub><mml:mo>,</mml:mo><mml:msub><mml:mi>r</mml:mi><mml:mrow><mml:mtext>out</mml:mtext></mml:mrow></mml:msub><mml:mo stretchy="false">)</mml:mo></mml:math> using a percentile radial-gradient operator with anatomical constraints, and then applies a crop-based polar transform to obtain a compact <mml:math id="mml-ieqn-2"><mml:mn>64</mml:mn><mml:mo>×</mml:mo><mml:mn>512</mml:mn></mml:math> unwrapped iris map. To better match polar geometry, we introduce <i>Learnable Polar Position Encoding</i> (LPPE) with separable radial–angular embeddings, where Fourier terms in the angular branch enforce continuity at <mml:math id="mml-ieqn-3"><mml:mi>θ</mml:mi><mml:mo>=</mml:mo><mml:mn>0</mml:mn><mml:mrow><mml:mo>/</mml:mo></mml:mrow><mml:mn>2</mml:mn><mml:mi>π</mml:mi></mml:math>. We further propose <i>Radial Stripe Window Attention</i> (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 <mml:math id="mml-ieqn-4"><mml:mi>P</mml:mi><mml:mo>×</mml:mo><mml:mi>K</mml:mi></mml:math> 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 <mml:math id="mml-ieqn-5"><mml:mn>3.5</mml:mn><mml:mo>×</mml:mo></mml:math> compared with a standard transformer baseline while maintaining competitive recognition accuracy.
KW  - Iris recognition; polar unwrapping; vision transformer; positional encoding; window attention; metric learning

DO  - 10.32604/cmes.2026.080616