Discriminative sparsity preserving projection via global constraint for unconstrained face recognition

The unconstrained face images collected in the real environments include many complicated and changeable interference factors, and sparsity preserving projections cannot well characterise the low-dimensional intrinsic structure embedded in the high-dimensional unconstrained face images, which is important for subsequent recognition task. To deal with this problem, in this study the authors propose a new dimensionality reduction method named as discriminative sparsity preserving projection via global constraint. It seeks an optimal sub-space in which the samples in intra-classes are as compact as possible, while the samples in inter-classes are as separable as possible by adopting the compactness constraint terms of reconstruction coefficients and the penalty terms of global distribution. Extensive experiments are carried out on Faces in Labeled the Wild database and PubFig database which are two representative unconstrained face sets, and the corresponding experimental results illustrate the effectiveness of the proposed method.