SaME: Sharpness-aware Matching Ensemble for Robust Palmprint Recognition.

Pose and illumination variations in unconstrained palmprint recognition cause critical problems in terms of region of interest (ROI) misalignment, defocus blur, and underexposured or overexposured imaging. However, most existing methods do not consider these quality factors when performing ROI matching; thus, palmprint recognition performance is sensitive to variations of palm poses and ambient light conditions. To address these problems, we propose the SaME strategy for robust contactless palmprint recognition. We have designed the sharpness-aware matching ensemble framework to exploit the advantages of different types of features while avoiding their limitations. First, we designed a quality scoring method based on an effective palmprint sharpness indicator. Second, a multi-feature extraction scheme was designed to take advantage of coarse-grained and fine-grained features. Finally, a quality-aware matching ensemble model is proposed to realize robust palmprint recognition. We conducted experiments on five contactless databases, and the results demonstrate that the proposed SaME framework can reduce the equal error rate (EER) significantly without complex ROI alignment. In addition, the EER value was less than 0.5% on the COEP x5 dataset that was generated with considerable quality variations.