Discriminant Neighborhood Preserving Embedding with L1-Norm Maximization
Discriminant neighborhood preserving embedding(DNPE)is a linear approach,which encodes discriminant information into the objective of neighborhood preserving embedding and improves its classification ability.However,the distance measure used in DNPE is L2-norm.As we all known,L2-norm is sensitive to abnormal noise.L2-norm based DNPE can be not robust to outliers in many practical applications.This paper presents an effective L1-norm maximization based DNPE algorithm(DNPE-L1)for biometric feature extraction.DNPE-L1 tries to maximize L1-norm based between-class dispersion and simultaneously tries to preserve the local manifold structure by reconstructing the objective of neighborhood preserving projection(NPE)based on L1-norm.Compared with some other related classical methods,extensive experimental results on face and palmprint recognition problem conducted on FERET face database,VALID face database and PolyU palmprint database indicate the effectiveness of the proposed DNPE-L1 algorithm.
CHEN Xi YAO Chuang ZHOU Zaihong SU Ming
School of Big Data and Computer Science,Guizhou Normal University,Guiyang,Guizhou,550025,China Information Center,National Natural Science Foundation of China,Beijing,100085,China School of Information Engineering,Guangdong Medical University,Dongguang,523808,China
国内会议
厦门
英文
172-180
2017-11-17（万方平台首次上网日期，不代表论文的发表时间）