Enhanced nuclear norm based matrix regression for occluded face recognition. (June 2022)

• Record Type: Journal Article
• Authors: Li, Qin; He, Huihui; Lai, Hong; Cai, Tie; Wang, Qianqian; Gao, QuanXue
• Is Part Of: Pattern recognition. Volume 126(2022)
• Publication Date: 2022-06
• Subjects: Face recognition -- Occluded image -- Nuclear norm -- Low-Rank; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2022.108585
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: Our model imposes a nuclear norm constraint on the reconstructed image suppressing the noise in training set by giving the contaminated training sample a small weight. Enforcing the reconstructed image to be low rank, our model endows small coefficients to the samples from the incorrect class and emphasize samples from the correct class. Our model avoids misclassification caused by collaborative representation consequently improves the discriminative ability of our model. Experiments show that the proposed model is more robust than NMR and other relate model in face recognition tasks with contiguous occlusions and illumination changes. Abstract: An effective approach for the task of face recognition is proposed in this paper, which formulates the problem as an enhanced nuclear norm based matrix regression model and explores the low-rank property of the reconstructed image. Previous works have already leveraged the nuclear norm to obtain a low-rank representation of the error image and get a promising recognition rate. Motivated by the low-rank property of the reconstructed image through theoretical observation, our model imposes the nuclear norm constraints not only on the representation residual but also on the reconstructed image. The proposed method preserves the 2D structural information of the error images and reconstructs images, which is significant for the face recognition tasks. To further improve the performance of the proposed model, we explore the impact of different regularization terms under various scenarios. Extensive experiments on several benchmark datasets show the efficacy of the proposed model especially in terms of robustness against contiguous occlusion and illumination changes, which achieves superior performance over the most competitive methods.