Biometric authentication systems are widely used in security applications as they use unique physiological or behavioral traits that are difficult to to duplicate or manipulate.
However, uni-modal biometric systems often suffer from noise sensitivity, poor-quality samples and vulnerability to various limitations that reduce authentication reliability. This
paper presents a multi-modal biometric authentication system that integrates fingerprint and retinal biometrics to improve identification accuracy and robustness. The proposed approach combines Local Binary Pattern (LBP)-based texture features and Convolutional Neural Network (CNN)-based deep features for both modalities, enabling the extraction of complementary local and high-level discriminative information. A systematic evaluation of
multiple score normalization methods and fusion strategies is carried out within a unified framework to analyze their impact on recognition performance. The normalized scores are
combined using a weighted score-level fusion approach to effectively utilize information from both modalities. Experimental evaluation on standard biometric datasets shows that the proposed system achieves a recognition accuracy of 98.81% with a low error rate of 1.1235%. The results indicate improved performance compared to uni-modal systems. The study also highlights the challenges associated with multi-modal dataset availability and
their impact on performance evaluation.