A Novel Feature Fusion Scheme for Human Recognition at a Distance.

Xianglei Xing,Kejun Wang,Xiaofei Yang,Tongchun Du

DOI: https://doi.org/10.1007/978-3-319-25417-3_64

2015-01-01

Abstract:Human identification at a distance remains a challenging problem. Two biometric sources that are available in such situations are gait and face. In this paper, we present a new approach that utilizes and integrates information from frontal gait and face at the feature level. A novel kernel coupled mapping method is introduced to project both the gait features and the face features into a unified subspace where the heterogeneous modalities are transformed into the homologous features naturally. Moreover, the proposed feature level fusion scheme is compared with the match score level fusion schemes (Sum, Max and Product rules) and two feature level fusion schemes. The experimental results demonstrate that the proposed feature level fusion scheme outperforms the match score level and the other two feature level fusion schemes.