A Biometric System with Hierarchical Feature-level Fusion

The paper approaches the feature-level fusion for biometric authentication, a big challenge for the actual biometric security systems design. A hierarchical inter-modal fusion method is proposed and evaluated for a reduced feature space. The feature generation and fusion are performed using regions of interest manually defined within the original images. The features are extracted using co-occurrence matrices, providing a common framework for feature generation among several human traits. The fusion is based on a functional combination of the features, an approach that is feasible if the feature vectors are homogeneous. The functional fusion avoids the concatenation that increases the feature space size, leading to curse of dimensionality and additional computational complexity costs.