Confidence Belief Function Weighted Parallel Rank-level Fusion for Face recognition

In this paper, proficient feature extraction techniques using efficient neural networks (NN) with evidence theory for face recognition are presented. This approach is established to reduce the computation periods required by these NN. Evidence theory based single or multi biometric fusion methods have established promising performance, but they cannot handle the uncertainty appropriately, suggesting that further improvement of the performance of single biometric authentication systems. Conventional ranking is upgraded, using some associations among the outputs (belief confidence factors) of a classifier. Then, the final result is achieved by fusing results from the combined classifier output (belief confidence factors) with evidence theory. The face database usually severely affected by various degradations such as, illumination, noise and pose variations etc. which affects the overall recognition accuracy. The outcome establishes that the proposed rank-level fusion method attains superior recognition accuracy than other feature extraction and other related rank level fusion approaches.