Novel rejection methods and fusion approaches for multi-biometric verification

This dissertation proposes methods and algorithms to improve the performance of biometric verification systems. It introduces a new rejection method, symmetric rejection method, for multi-stage biometric verification. The symmetric rejection method significantly improves the performance over the state of the art rejection methods and controls the genuine reject rate which has not been specifically addressed in earlier studies. The dissertation also proposes a new fusion framework for multi-biometric verification systems, which achieves accuracy higher than parallel fusion framework, and provides convenience to genuine users. In addition, it proposes a framework consisting of impostor score based normalization, impostor score based rejection, and fusion to lower the verification errors of continuous keystroke verification with weak templates. It introduces a new formulation to incorporate the reject option in verification with weak templates and develops a new impostor score based rejection method called Order Statistic rejection method . Results show that the proposed framework in conjunction with the Order Statistic rejection method significantly reduces the equal error rates of continuous keystroke verification with weak templates