摘要
稀疏表示系数包含较强的鉴别信息,稀疏保持投影(Sparsity Preserving Projections,SPP)利用稀疏表示系数进行特征提取.本文通过核方法获取高维特征空间的核稀疏表示系数,并利用核稀疏表示系数构造邻接矩阵,提出核稀疏保持投影(Kernel Sparsity Preserving Projections,KSPP).核稀疏表示系数比稀疏表示系数包含更强的鉴别信息,因此KSPP可以比SPP提取更有效的鉴别特征.在多个数据库上的生物特征识别实验,KSPP都取得了不错的实验结果.
Sparse representation coefficient contains strong discriminant information and sparsity preserving projections ex- tracts features by sparse representation coefficient. This paper obtains kernel sparse representation coefficient in the high dimensional space by kernel method and use kernel sparse representation coefficient to construct adjacency matrix, then propose kernel sparsity preserving projections. Kernel sparse representation coefficient contains stronger discriminant information than sparse representation coefficient; therefore, KSPP could extract more efficient features than SPP. KSPP achieves good results in biometrics experiments of several databases.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2013年第4期639-645,共7页
Acta Electronica Sinica
基金
国家自然科学基金青年基金(No.61005008)
关键词
稀疏表示
邻接矩阵
稀疏保持投影
核方法
sparse representation
adjacency matrix
sparsity preserving projections
kernel method