Robust periocular authentication in HMD environments using multi‐template fusion strategies

Abstract Periocular authentication in head‐mounted display (HMD) environments is challenged by substantial intra‐class variability caused by gaze shifts, blinking, facial expressions, and HMD‐specific imaging artifacts. Such variability limits the reliability of single‐template verification. We present a robust periocular authentication framework and systematically compare multi‐template selection and fusion strategies to address this problem. Our method integrates eye‐region segmentation, gaze estimation, and geometric normalization to support reliable matching in HMD settings. We evaluate basic fusion rules and weighted fusion using template quality and query‐template similarity. Experiments on the AffectiVR dataset show that multi‐template strategies substantially improve upon the single‐template baseline, with the best weighted fusion strategy achieving 1.22% equal error rate (EER) with 16 templates, a 71% relative reduction from the baseline EER of 4.19%. These results demonstrate the effectiveness of multi‐template strategies for robust HMD periocular authentication.