CAJ | 학술논문

稀疏保留投影( SPP)是一种保留样本间的稀疏重构关系的特征提取方法。但是根据流形学习理论,考虑局部流形结构比考虑全局欧氏结构更重要。此外,SPP得到的不是一组正交的投影向量,特征间存在冗余信息。为解决该问题,文中提出一种改进的稀疏保留投影算法,在SPP中引入有监督的流形学习,使得所得投影空间正交,并用迭代的方式求解最优投影变换,称为基于流形学习的迭代正交稀疏保留鉴别分析( MLIOSDA)。同时提出一种终止准则终止迭代。在CAS-PEAL人脸数据库和PolyU掌纹数据库的实验结果表明,文中提出的方法与一些相关方法相比有效地提高了识别结果。
희소보류투영( SPP)시일충보류양본간적희소중구관계적특정제취방법。단시근거류형학습이론,고필국부류형결구비고필전국구씨결구경중요。차외,SPP득도적불시일조정교적투영향량,특정간존재용여신식。위해결해문제,문중제출일충개진적희소보류투영산법,재SPP중인입유감독적류형학습,사득소득투영공간정교,병용질대적방식구해최우투영변환,칭위기우류형학습적질대정교희소보류감별분석( MLIOSDA)。동시제출일충종지준칙종지질대。재CAS-PEAL인검수거고화PolyU장문수거고적실험결과표명,문중제출적방법여일사상관방법상비유효지제고료식별결과。
Sparsity Preserving Projections ( SPP) is an effective feature extraction method,which can preserve the sparse reconstruction re-lations among samples. However,according to the manifold learning theory,the local manifold structure of samples is more important than the global Euclidean structure of samples. SPP cannot get a set of orthogonal projection vectors,and thus there exists redundant informa-tion among the obtained features. To address these problems of SPP,propose a novel approach called Manifold Learning based Iterative Orthogonal Sparsity preserving Discriminant Analysis ( MLIOSDA) ,which introduces the idea of manifold learning into SPP and obtains orthogonal projection space. Obtain optimal projection vectors in an iterative manner. Also provide a terminating criterion to finish the it-eration. Experimental results on CAS-PEAL and PolyU databases demonstrate that the proposed approach can effectively improve the rec-ognition results compared with some related methods.