<p>Biometric identification is one of the secure solutions to identify a person’s identity and protect information security. However, a single biometric system cannot handle diverse authentication scenarios due to the presence of noise and higher implementation costs. To address this issue, the given paper introduces a secure multimodal biometric authentication framework that makes use of fused convolutional neural network (FCNN) and hash-based cryptography. FCNN performs extraction, fusion, and classification of features. Feature extraction involves pre-processing of images followed by convolution of biometrics through a pooling layer. The feature layer fusion approach considers image features as input to the self-attention mechanism to achieve corresponding weights. The classification part is represented in the fully connected (FC) layer. The hash function operates on the pre-processed biometric data, resulting in an output as a binary string of a given length, commonly called a hash value. The performance of the proposed framework is validated and compared with existing studies based on recognition accuracy (RA %) and false positive rate (FPR %). The results show superior performance, thereby enhancing verification and security of the system.</p><p>.</p>