基于低秩表示线性保持投影的特征提取算法

为了在特征提取过程中保持数据低秩特性不变,提出了一种基于低秩表示的线性保持投影算法用于维数约简。它能够使降维后的低维空间中的数据依旧较好地保持在原始高维空间中的低秩特性,准确地学习出数据的低维子空间。通过构建两个不同的低秩表示模型来 揭示两种不同结构特性的低秩权重,然后以保持数据的这两个低秩权重关系为目的来求解高维数据的低维空间。 在ORL库和Yale库人脸库上的实验结果证明,该算法比传统的特征提取方法更有效。

Feature Extraction Based on Low Rank Representation Linear Preserving Projections

For preserving the low rank properties the same,we proposed an algorithm,called linear preserving projection based on low rank representations (LLRLPP),to reduce the dimension of data.It can preserve the low rank properties of the original data space in the resulting low dimensional embedding subspace and correctly learn the low-dimensional subspace.Through constructing two different low rank representation model,the low rank weights of representing different structural characteristics are revealed.Then the low-dimensional subspace of the original high-dimensional data is obtained by preserving such low rank weight relationship.The effectiveness of the proposed method is verified on two face databases(ORL,Yale) with the traditional algorithms.