基于鉴别混合结构保持投影的人脸识别算法  

Face Recognition Algorithm Based on Discriminative Hybrid Structure Preserving Projection

作　　者：王成国 胡西川[1] WANG Cheng-guo;HU Xi-chuan(College of Information Engineering,Shanghai Maritime University,Shanghai 201306)

机构地区：[1]上海海事大学信息工程学院,上海201306

出　　处：《现代计算机（中旬刊）》2018年第8期61-65,共5页Modern Computer

摘　　要：基于混合结构保持投影的降维算法(SNPP)将数据的近邻表示和稀疏表示深度融合成一种混合结构,可以更好地保持原数据特征。鉴别混合结构保持投影(DSNPP)在SNPP的基础上,再加上鉴别信息并结合改进最大余量原则(MMMC),可以使经算法处理过得数据有更强的分类性能。经过在PIE和AR人脸库上的实验,验证DSNPP算法的有效性。The linear dimensionality reduction algorithm based on hybrid structure-preserving（SNPP） projection fusions the near neighbor representation and sparse representation of data into a hybrid structure, which can better preserve the original data features. This algorithm adds authentication information to SNPP and incorporates the improved Maximum Margin Principle（MMMC）, which helps to make full use of the data authentication information. It can make the algorithm have stronger classification ability. The experiments on PIE and AR face database verify the effectiveness of DSNPP algorithm.

关 键 词：稀疏保持 近邻保持 最大余量准则 鉴别信息 降维 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]