A unimodal fusion framework utilizing multiple enrolment images

One of the main purposes of using multiple enrolment images in a biometric recognition system is that if the input image is noisy and/or distorted in some way or is deformed during capturing stage, the recognition system may still be able to match with the genuine enrolment image providing more than one option is available. This approach, as used in conventional unimodal recognition systems, attempts to match the input image with multiple enrolment images for each user and select the highest matching score as the correct matching result. However, there exist some inter-dependencies between the data and matching scores which can be exploited not only to enhance the recognition results but also to achieve a better consistency of matching. In this paper we propose an approach to compute and utilize the relationship between the matching scores and hence exploits the computed inter-dependency measures to (i) achieve an improved and stable recognition results and (ii) provide a strong quality measure that can be employed not only to reduce Error Rates but also to act as a parameter that can be used to set weights for multimodal fusion base recognition systems. To demonstrate the effectiveness of the proposed methods, extensive experiments were carried out on fingerprint and iris data using widely used databases.