稀疏正交普鲁克回归处理跨姿态人脸识别问题

正交普鲁克分析是 一种常用的处理矩阵近似问题的技术。最近,该技术被引入到正交普鲁克回归模型中来处理人脸姿态识别问题并取得了不错的效果。然而,这个模型对残差项使用了矩阵F范数约束,使得模型对于一些噪声(比如光照)非常敏感。为解决该问题,用更加鲁棒的1范约束替代原始的矩阵F范数约束,提出稀疏正交普鲁克回归模型。该模型可以由一个有效的交替迭代算法解决。在几个流行的人脸数据库上做了验证实验,实验结果证明该模型可以有效地处理人脸姿态变化。

Sparse Orthogonal Procrustes Problem Based Regression for Face Recognition with Pose Variations

Orthogonal Procrustes problem (OPP) is a popular technique to deal with matrix approximation problem.Recently,OPP was introduced into a regression model named orthogonal Procrustes problem based regression (OPPR) to handle facial pose variations and achieved interesting results.However,OPPR performs F-norm constraint on the error term,which makes the model sensitive to the noises (i.e.,illumination variations).To address this problem,in this paper,the F-norm constraint was replaced by the L₁-norm constraint and the sparse orthogonal Procrustes problem based regression (SOPPR) model was proposed,which is more robust.The proposed model was then solved by an efficient alternating iterative algorithm.Experimental results on public face databases demonstrate the effectiveness of the proposed model for handling facial pose variations.