Eigen-based binary feature amalgamation in multimodal biometrics

In this paper, a quantised eigen analysis (QEA) for the extracted features is proposed and an associated eigen-based binary feature amalgamation (EBFA) based on QEA is developed for feature fusion in multimodal biometrics. As opposed to feature combination, EBFA projects heterogeneous features onto the projection kernel and uses only the sign parts to encode the features as bit strings to maximise its expression rather than directly combine them. Thus, the feature codes can be simply concatenated or compared by XOR bit-wise operation into a serial or parallel amalgamated feature vector. To evaluate the performance of EBFA, a series of experiments are performed on multiple biometric modalities, including face, palm-print and iris. The experimental results show that the proposed binary feature amalgamation scheme at feature-level is superior to some other feature fusion methods and score-level methods in terms of multimodal recognition accuracy performance.