Biased subspace learning for misalignment-robust facial expression recognition

In this paper, we propose a biased subspace learning approach for misalignment-robust facial expression recognition. While a variety of facial expression recognition methods have been proposed in the literature, most of them only work well when face images are well registered and aligned. In many practical applications such as human robot interaction and visual surveillance, it is still challenging to obtain well-aligned face images for facial expression recognition due to currently imperfect computer vision techniques, especially under uncontrolled conditions. Motivated by the fact that interclass facial images with small differences are more easily mis-classified than those with large differences, we propose a biased linear discriminant analysis (BLDA) method by imposing large penalties on interclass samples with small differences and small penalties on those samples with large differences simultaneously, so that discriminative features can be better extracted for recognition. Moreover, we generate more virtually misaligned facial expression samples and assign different weights to them according to their occurrence probabilities in the testing phase to learn a weighted BLDA (WBLDA) feature space to extract misalignment-robust discriminative features for recognition. To better exploit the geometrical information of face samples, we propose a weighted biased margin Fisher analysis (WBMFA) method by employing a graph embedding criterion to extract discriminative information, so that the assumption of the Gaussian distribution of samples is not necessary. Experimental results on two widely used face databases are presented to show the efficacy of the proposed methods.