Abstract:
Multimodal biometrics combines a diversity of biological traits in an attempt to produce a notable influence on
identification performance. In recent years, multimodal biometric recognition using machine learning algorithms has received
considerable attention. This study proposes a novel multi modal biometrics recognition method based on Multi-scale Geometric
Curvelet (MGC) and Minkowski distance factor models. The new method is termed, Geometric Curvelet and Minkowski Multimodal
Biometric Recognition (GC-MMBR), and works as follows. First, an intrinsic representation of multimodal features namely
fingerprint, face and iris traits) using Rationalized AdaBoost is learnt. Second, a MGC Feature Extraction model is applied to the
resultant preprocessed features, to extract intrinsic curve features. Finally, the reconstructed, extracted intrinsic features are used as
input to a Minkowski distance-based biometric recognition approach. When compared with existing methodologies, the proposed
multimodal biometric recognition algorithm is proven to perform well in terms of recognition rate. Specifically, comparative
evaluation using the benchmark, CASIA Biometric Ideal Test Dataset, shows our proposed GC-MMBR achieves 35% overall
recognition rate, out-performing existing methods. Comparative FINDINGS further PROVED the ability of proposed GC-MMBR to
considerably reduce computational complexity and false acceptance rate. Thus, we conclude our proposed method can provide
benchmarking performance for conventional biometric recognition methods.