Multimodal key-binding biocrypto-system using least-square polynomial curve-fitting based new feature-level fusion method

Abstract

In last few years, many works has been proposed using multimodal biometric system because of its high performance as compare to unimodal biometric systems. Most of the Multimodal Biocrypto-System (MBS) have been previously proposed to securely share secret-key over the network, but these systems uses complex signal processing techniques like DFT, SVM, neural network etc. based fusion techniques and relatively low performance. Therefore, we propose simple and effective but mathematically irreversible statistically based new feature level fusion technique using Least-Square Polynomial Curve-Fitting (LPC) for the proposed efficient Multimodal Key-Binding Biocrypto-System (MKBB). We validate the effectiveness of proposed technique over the fuzzy vault scheme using biometrics fingerprint and iris datasets. This proposed system is implemented to protect the user’s cryptographic secret-key and effectively remove the use of public key infrastructure (PKI) system because of its complex certification issuing and distributing management costs, and centralized structure which uses convention network system and shows single point of failure. We also evaluate the overall performance system to successfully retrieval of key with the help of AES-256 algorithm to perform encryption and decryption. The experimentations is done using fingerprint FVC2002DB_1 and Iris CASIA-IrisV1 datasets. The system gives the 99.96% of accuracy, with 99.98% of GAR and 0% FMR.