基于多稀疏分布特征和最近邻分类的物体识别方法  

Object recognition method based on multi-sparse distribution features and nearest neighbor classification

作　　者：孙利娟[1,2] 张继栋[3] 杨新锋[4] Sun Lijuan;Zhang Jidong;Yang Xinfeng(School of Computer Science, Wuhan University, Wuhan 430012, China;Dept, of Information Electronic, Kaifeng Institute of Education,Kaifeng Henan 475000 , China;Dept, of Computer, Zhengzhou Institute of Finance & Economics , Zhengzhou 450007 , China;School of Computer & Information Engineering, Nanyang Institute of Technology, Nanyang Henan 473004 , China)

机构地区：[1]武汉大学计算机学院,武汉430072 [2]开封教育学院信息电子系,河南开封475000 [3]郑州财经学院计算机系,郑州450007 [4]南阳理工学院计算机与信息工程学院,河南南阳473004

出　　处：《计算机应用研究》2016年第10期3156-3159,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(41001292);河南省重点科技攻关计划资助项目(122102210563;132102210215)

摘　　要：为提高物体识别性能,提出了一种基于多稀疏分布特征和最近邻分类的目标识别方法。提取图像的梯度模值和方向特征,构建梯度模值和方向图像,分别对灰度图像、梯度模值图像和梯度方向图像进行稀疏表示,提取稀疏分布特征,得到融合后的多稀疏分布特征,再依据最近邻分类方法进行特征分类,实现物体识别。通过在国际公认的COIL-100和PVOC-2007两个公共测试数据集下进行对比实验,对提出方法的参数选择、鲁棒性和识别性能进行综合评价。实验结果表明,采用提出的方法进行物体识别的识别率高于目前经典的SIFT、SURF和ORB方法,是一种有效的物体识别方法。In order to im prove the performance of object recognition , this paper proposed an object recognition method basedon multi-sparse distribution features and nearest neighbor classificatio n. I t extracted the features o f gradient m agnitude and d irection o f im ag e, and constructed gradient m agnitude image and gradien t d ire ctio n image. T h e n , it executed sparse representation on gray im a g e , gradient m agnitude image and gradient d ire c tio n image re sp e ctive ly, to extract sparse d is trib u tio n fe a tu re s,and obtained the m u lti-sparse d is trib u tio n features. F in a lly , it classified the features o f d iffe re n t objects according to nearestneighbor classificatio n m e thod , to realize ob je ct recognition. I t im plem ented experim ents on two in te rn a tio n a l common datasetin c lu d in g COIL-100 and PVOC-2007 , and evaluated com prehensively o f the parameters selectio n, robustness and recognitionperform ance o f the new m ethod. The results show that the new m ethod has h igh er accuracy than three classical methods in cluding S IFT ,SURF and ORB on object recognition , and is avalid object recognition method.

关 键 词：物体识别 稀疏表示 最近邻距离 梯度 街区距离 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]