INTERNATIONAL JOURNAL FOR INNOVATIVE RESEARCH IN MULTIDISCIPLINARY FIELD

Multimodal biometric systems improve recognition reliability by combining multiple biometric traits. However, traditional score-level fusion techniques assign fixed weights to modalities and ignore sample quality variations, resulting in performance degradation under poor acquisition conditions. This paper proposes a quality-aware score-level fusion framework for face and palm vein biometrics using Mobile Vision Transformer (MobileViT) for feature extraction. In practical environments, however, biometric data often suffer from quality degradation due to illumination changes, motion blur, occlusion, sensor noise, or physiological variations. Ignoring such quality variations can significantly reduce system performance and increase false acceptance and false rejection rates. The proposed system dynamically adjusts fusion weights based on modality quality measures before computing the final decision score. Experimental results demonstrate that the quality-aware approach consistently outperforms conventional fusion techniques, achieving higher true positive rates and improved area under the ROC curve (AUC). The proposed method attains superior Area Under Curve (AUC) values and reduced Equal Error Rate (EER). The adaptive weighting mechanism ensures that higher-quality samples contribute more significantly to the final decision score, while lower-quality inputs are proportionally suppressed. The findings confirm that adaptive quality weighting enhances robustness and overall recognition performance.