摘要
在空间金字塔词袋模型的基础上,针对其空间信息利用不足的问题,本文先计算图像中每一个字典向量的相对位置分布来提取出局部位置特征.然后,用非下采样轮廓波变换和线性判别分析来生成图像的全局轮廓特征.最后,通过局部位置特征与全局轮廓特征相结合的方式提高空间信息利用率,从而提高场景和物体图像分类正确率.为了检验方法的可行性,本文分别在数据库Caltech 101、MSRC和15 Scene上进行实验.实验结果证明,本文提出的方法进一步利用了空间信息,从而提高了分类正确率.
Based on Spatial Pyramid Matching method,aiming at the insufficient utilization of spatial information,firstly,local position feature is extracted by computing relative position distribution of each dictionary vector in image. Then,global contour feature is generated through Nonsubsampled Contourlet Transform and Linear Discriminant Analysis. Finally,Spatial information is enhanced by combining local position feature with global contour feature,which consequently improves the accuracy of scene and object classification. Extensive experiments are performed on Caltech 101,MSRC and 15 Scene datasets respectively. The experimental results showthat the proposed method further utilizes the spatial information,and thus improves the accuracy of image classification.
作者
李雅倩
吴超
李海滨
刘彬
LI Ya-qian;WU Chao;LI Hai-bin;LIU Bin(Key Lab of Industrial Computer Control Engineering of Hebei Province,Yanshan University,Qinhuangdao,Hebei 066004,China;Institute of Information Science and Engineering,Yanshan University,Qinhuangdao,Hebei 066004,China)
出处
《电子学报》
EI
CAS
CSCD
北大核心
2018年第7期1726-1731,共6页
Acta Electronica Sinica
基金
河北省自然科学基金(No.F2015203212)
关键词
图像分类
词袋模型
局部位置特征
非下采样轮廓波变换
全局轮廓特征
image classification
bag of words
local position feature
nonsubsampled contourlet transform
global contour feature