基于SPM-IBOVW模型的自然场景识别  

Natural Scene Recognition Based on Spatial Pyramid Integrated Bag-of-Visual-Words Model

作　　者：章海兵[1,2] 刘士荣[1] 张波涛[1] 王坚[1] 

机构地区：[1]杭州电子科技大学自动化学院,检测仪表与自动化系统集成技术教育部工程研究中心,杭州310018 [2]浙江国自机器人技术有限公司,杭州310053

出　　处：《上海交通大学学报》2016年第6期902-909,共8页Journal of Shanghai Jiaotong University

基　　金：国家自然科学基金项目(61175093,61503108);浙江省自然科学基金项目(LQ14F030012)资助

摘　　要：提出一种基于空间金字塔匹配的集成化视觉词典(SPM-IBOVW)模型的自然场景识别方法.对各个类别的特征集合分别聚类形成各类别视觉词包,并连接成为集成化视觉词包(IVV);采用一种新的5层空间金字塔划分方法,相比经典划分方法在降低算法复杂度的同时,提高了场景识别率,并通过15类通用复杂场景的识别实验验证其有效性.结果表明,所提出的方法对复杂场景的识别率达到83.17%.A method based on spatial pyramid matching-integrated BOVW（SPM-IBOVW）was proposed for natural scene recognition.Each set of features were clustered to form various types of vocabularies,which were connected into an integrated visual vocabulary.A novel division method with five levels of spatial pyramid was presented.Compared with the classical division method,the proposed division method can reduce the algorithm complexity and improve the recognition rate of scene.Experimental results show that the proposed method has an average recognition rate of 83.17%in the large common dataset of 15 categories of complex scenes.

关 键 词：视觉单词 空间金字塔匹配 集成化视觉词包模型 集成化视觉词典 场景识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]