An improved and robust fusion framework for soft biometric traits

The use of soft biometrics based recognition systems has been on the rise due to their wide range of potential applications and the numerous benefits incurred over traditional biometric systems. In these systems, soft biometric traits are combined with primary biometric modalities in a multimodal fusion environment. The most popular and widely used technique for implementing this fusion process is the probabilistic framework suggested by Jain et.al. In our work we have significantly improved upon this basic construction by not only making it more robust regarding the nature of soft biometric trait values, but also incorporating time variant soft biometric information in it. Our proposed design makes the basic framework more suited for handling real life scenarios since it does not assume any predefined distribution in the underlying data. Moreover, the performance of the overall framework significantly increases due to the enhancement in the matching accuracy of the soft biometric data. The results obtained from experiments confirm our ideas.