Single Sample Face Recognition with Gabor Feature based Linear Regression

By constructing a linear model representing using the auxiliary intra-personal variations, the adaptive linear regression classifier(ALRC) successfully popularized the linear regression classifier(LRC) to the single sample per person(SSPP) scenario. However, ALRC simply uses original face images to constitute the feature space, which is not robust enough against the variations of probe images, and leads to the algorithm computationally very expensive. To address the two problems, in this paper, the image Gabor features is used for ALRC scheme. By using Gabor kernels based feature space, the abilities of ALRC against variations can be improved. Meanwhile, principal component analysis(PCA) is implemented in the feature extracted stage. Experiments on representative face databases show that Gabor-feature based ALRC can achieve better recognition performance than original ALRC, and reduce much computational burden compared with later.