Matching NIR face to VIS face using multi-feature based MSDA

Visual and near infrared (VIS-NIR) face image matching, which is also called cross-spectral face matching in heterogeneous face recognition, is important for security application. However, most existing methods perform poorly in this scenario because of the cross-modality appearance differences. To address this problem, we propose a new method named multi-feature based Multi-view Smooth Discriminant Analysis (MSDA) in this paper. The proposed method involves three kinds of local feature descriptors (i.e., Histogram of Oriented Gradient, HOG; Local Triplet Pattern, LTP; Scale-invariant feature transform, SIFT). In addition, MSDA is formulated for finding a multi-view learning based common discriminative feature space and it can utilize the underlying relationship of features from different modalities. Extensive experiments demonstrate the superiority of the new proposed multi-feature based MSDA approach for VIS-NIR face matching.