Evaluation of fusion at different levels for face recognition
Prachi Punyani, Ashwani Kumar · 2017
Biometrie systems face several limitations like low accuracy, less robustness, low applicability and non-universality which can be minimized by the adoption of Fusion at different levels of Biometric systems. Fusion can be applied at sensor level, features level, score level as well as decision level. In this paper, we have compared the performance of a Face Recognition system without fusion and with fusion applied at different levels. Weighted majority voting, Concatenation, Borda Count and Weighted averaging have been applied at sensor, features, score and decision levels respectively for comparison of Recognition rate results. For effective comparison study, all experiments have been performed on face images using two different publically available FERET database and MORPH2 database. It has been reported that Fusion always improves the performance of the system. Moreover, Score level fusion improves the accuracy of the system to the maximum rate followed by features level fusion. Sensor level fusion is the next preference and the least preferred fusion is at Decision level having minimum recognition accuracy on both FERET and MORPH2 database.