Combining space-based Gabor features for face recognition

We propose a face recognition strategy combining various discriminating Gabor features in multi-scales and multi-orientations. A bank of well-chosen Gabor filters is applied on the image to construct a group of feature vectors, and then the Null Space-based LDA (NLDA) is performed simultaneously on each orientation channel and the original image to give 5 component classifier outputs, which are then combined to increase the final recognition rate. Experimental results on the FERET database demonstrate the effectiveness and flexibility of our proposed method.