基于流形学习的正交稀疏保留投影

稀疏保留投影通过保留样本之间的全局稀疏重构关系来进行特征提取,获得了良好的分类效果。但是,稀疏保留投影得到的投影变换通常不是正交的,而且在实际应用中,正交性一直被认为有利于提高鉴别能力。另外,根据流形学习理论,局部流形结构比全局欧式结构更重要。因此,文中在稀疏保留投影中引入了流形结构保留和正交投影,提出了整体正交流形稀疏保留投影（HOMSPP）和迭代正交流形稀疏保留投影（IOMSPP）两种实现算法来实现人脸和掌纹图像的特征提取。

Orthogonal Sparsity Preserving Projections Based on Manifold Learning

Sparsity Preserving Projections （ SPP） extracts features by preserving the global sparse reconstruction relations among samples, which achieves favorable classification results. However,the obtained transformation of SPP usually is not orthogonal,while in real appli-cations,orthogonality is advantageous for classification in many scenarios. Besides,according to the manifold learning theory,local mani-fold structure is more important than global Euclidean structure. Therefore,in this paper,introduce manifold preserving and orthogonal transformation into SPP,and propose two novel approaches for face and palmprint image feature extraction,which are Holistic Orthogonal Manifold and Sparsity Preserving Projections （ HOMSPP） and Iterative Orthogonal Manifold and Sparsity Preserving Projections （ IOM-SPP） .