A secure blockchain-enabled authentication and authorisation framework for cloud environments using distributed biometric verification

Centralised cloud authentication and authorisation commonly rely on a trusted server for identity, policy enforcement, and audit logging, which can increase the impact of compromise and complicate tamper-evident accountability. This paper presents a decentralised biometric authentication and authorisation framework that integrates fingerprint minutiae verification with Ethereum smart contracts and IPFS-based off-chain storage. Biometric templates are processed off-chain using a NIST NBIS-based pipeline ( MINDTCT minutiae extraction and Bozorth3 matching) and protected via SHA-256 hashing, zlib compression, and authenticated encryption (Fernet), while the blockchain stores only integrity metadata (hashes and IPFS content identifiers) and authorisation state. To mitigate replay attacks without introducing additional session-creation transactions, the framework uses a domain-separated request-binding hash ( tokenHash ) over the caller, nonce, timestamp, chain id, and contract address, validated on-chain together with one-time nonce consumption. A prototype was implemented using Solidity on a local Ethereum-compatible test network (Ganache), with Python/Web3 tooling and an IPFS-compatible content-addressed storage interface. Biometric performance is reported using standard metrics (FAR/FRR/EER and ROC/DET curves) on SOCOFing and FVC2002/FVC2004 benchmarks. In the core blockchain experiment, the proposed single-call attendance workflow required 74,773 gas while adding nonce-based replay protection and auditable authorisation checks on top of CID/hash anchoring (the separate literature benchmark reports end-to-end call-path gas under a shared harness). Compression reduces encrypted-template storage by approximately 76–84% across tested dataset sizes. Security analysis and attack simulation show that replay attempts (nonce/token reuse), unauthorised access, and off-chain payload tampering are rejected under the stated threat model.