SRC-ODP:面向稀疏表示分类器的正交鉴别投影  

SRC-ODP: sparse representation classifier oriented orthogonal discriminative projection

作　　者：赵家成[1] 张国庆[1] 孙怀江[1] Zhao Jiacheng;Zhang Guoqing;Sun Huaijiang(School of Computer Science & Engineering, Nanjing University of Science & Technology, Nanjing 210094 , China)

机构地区：[1]南京理工大学计算机科学与工程学院,南京210094

出　　处：《计算机应用研究》2016年第10期3165-3168,共4页Application Research of Computers

摘　　要：稀疏表示分类方法(sparse representation-based classifier,SRC)在模式识别领域展现了巨大的潜力。基于稀疏表示分类的鉴别投影(SRC steered discriminative projection,SRC-DP)则是建立在SRC分类准则基础上的降维方法,其在投影空间中最大化类间重构误差与类内重构误差的比值。针对SRC-DP中提取的特征之间具有冗余信息,从而影响其鉴别能力的问题,提出SRC-ODP(SRC oriented orthogonal discriminative projection)方法,利用投影矩阵的正交约束取代SRC-DP中的约束条件,其优越性为:a)正交投影矩阵具有更高的特征提取效率;b)所提取的特征具有更强的鉴别能力。在AR和Extended Yale B数据库上的实验表明,该方法可以使SRC达到更好的分类结果。SRC has shown great po tentia l in pattern recognition. S R C-D P is a d im e n sio n a lity re duction m ethod based on SRC.I t maxim izes the ratio o f between-class reconstruction re sid ual to w ith in -cla ss reconstruction resid ual in the p ro je ctio n space. Toreduce the in flu e n ce o f the re dundant info rm atio n between features in S RC-DP , this paper proposed a new m ethod ca lle d SRCoriented orthogonal d iscrim in a tive p rojection (SRC-ODP) and em ployed orthogonal con straint on the projection m a trix , instead of the constraint condition in SRC-DP. The m ethod has two advantages: a )Projection matrix with orthogonal constraint is more efficient in feature extraction; b )Features extracted by this method have higher power in discrimination . Experim ents on theA Rand extend Yale B databases dem onstrate that th is m ethod is more effective based on the SRC.

关 键 词：稀疏表示分类 正交鉴别投影 特征提取 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]