Two-dimensional Ridge Regression (2DRR) for Face Recognition

摘要

In this paper, we propose a novel method called two-dimensional ridge regression (2DRR) for face recognition, which is based on 2D image matrices rather than ID vectors as conventional ridge regression does. Due to the fact that regular simplex vertices are highly symmetric and well separated, in ID case, image vectors from each individual are projected around a regular simplex vertex using ridge regression, which guarantees the separability in a low dimensional space. However, the dimension of image vector is usually very high which makes the computational cost very expensive. 2DR.R overcomes the limitations of conventional ridge regression. Experiment results on the ORL and AR face image databases show the effectiveness and robustness of the proposed method.