Skip to Main Content
In this paper we present the face recognition method using feature-level fusion where the infrared (IR) and visible face images are fused at transformed domain. The main contribution of this work is the fusion scheme performed at nonlinear transformed domain. We examine two nonlinear face subspaces: Kernel Principle Component Analysis (KPCA) and Kernel Fisherpsilas Linear Discriminant Analysis (KFLD). The IR and visible feature components are extracted by the kernel methods and then concatenated using GA as a tool for optimal fusion strategy. We compare the recognition performance of the fusion scheme in the kernel based subspaces with the single modality of IR and visible based recognition. The experimental results show that the combination of fusion scheme based on real-valued GA and KFLD method appears to be the best at simultaneously handling the drawback of single modality face recognition and the weakness of linear based face subspaces.