Secure Cloud Storage Using Multi-Modal Biometric Cryptosystem: A Deep Learning-Based Key Binding Approach

Cloud computing has transformed data storage but presents security challenges, especially in authentication. Traditional passwords are vulnerable to attacks, while biometric authentication offers an alternative using fingerprints and facial recognition. However, biometric templates cannot be revoked if compromised. To address this, biometric cryptosystems integrate authentication with cryptography, though existing methods face computational and security challenges. This paper proposes a secure biometric cryptosystem for cloud storage using deep learning for biometric recognition and key binding techniques like fuzzy extractors and SHA-256 hashing. The system follows three phases: enrollment (feature extraction via YOLOv8 and DeepFace-VGG, fingerprint hashing with SHA-256, and key binding using fuzzy extractors), verification (cosine similarity for biometric matching), and encryption (AES-256 for secure storage). Experimental results show high authentication accuracy (mAP of 0.984 at 50% IoU), with FAR of 0%, FRR of 0.93%, and GAR of 99.07%. The comparative analysis highlights DeepFace-VGG’s effectiveness in feature extraction, SHA-256’s secure key generation, and fuzzy extractors' reliability in key reconstruction. The system resists replay attacks and biometric fluctuations while maintaining usability. Findings confirm that biometric authentication with cryptographic key binding enhances cloud security. Future research should explore scalability, multimodal biometrics, and advanced security methods like blockchain and homomorphic encryption