RSOTP子空间学习方法

A Reversible and Stable Orthogonal Tensor Projection Method in Subspace Learning

为增强张量子空间学习的稳定性、可重现性,及其分类能力,提出了一种鲁棒的子空间学习方法——RSOTP。该方法以MFA为鉴别准则,采用稳定的初始化向量和交替形式的正交来去除了权重因子,从而使整个学习过程不仅具有鲁棒性、可重现性,而且具有较强的分类能力,该方法在大小型人脸数据库上进行的实验均取得了较高的识别率。

In this paper we proposed a MFA based RSOTP subspace learning method. By employing stable discriminant vector in initialization and using an alternative orthogonal selection and throwing off weight matrix, this approach enhances the robustness of the learning process and strengthens the classification ability, and achieves better recognition performance on both large and small face datasets than conventional techniques.